Dissertations / Theses on the topic 'Trait estimation'

Doctor of Philosophy<br>Department of Statistics<br>Weixing Song<br>In this dissertation, we proposed a new robust estimation procedure for two multivariate mixture regression models and applied this novel method to functional mapping of dynamic traits. In the first part, a robust estimation procedure for the mixture of classical multivariate linear regression models is discussed by assuming that the error terms follow a multivariate Laplace distribution. An EM algorithm is developed based on the fact that the multivariate Laplace distribution is a scale mixture of the multivariate standard normal distribution. The performance of the proposed algorithm is thoroughly evaluated by some simulation and comparison studies. In the second part, the similar idea is extended to the mixture of linear mixed regression models by assuming that the random effect and the regression error jointly follow a multivariate Laplace distribution. Compared with the existing robust t procedure in the literature, simulation studies indicate that the finite sample performance of the proposed estimation procedure outperforms or is at least comparable to the robust t procedure. Comparing to t procedure, there is no need to determine the degrees of freedom, so the new robust estimation procedure is computationally more efficient than the robust t procedure. The ascent property for both EM algorithms are also proved. In the third part, the proposed robust method is applied to identify quantitative trait loci (QTL) underlying a functional mapping framework with dynamic traits of agricultural or biomedical interest. A robust multivariate Laplace mapping framework was proposed to replace the normality assumption. Simulation studies show the proposed method is comparable to the robust multivariate t-distribution developed in literature and outperforms the normal procedure. As an illustration, the proposed method is also applied to a real data set.

In a study looking for risk factors of atherosclerosis in families with combined hyperlipidemia and hypertension, clinical and biochemical data of 1,149 persons were analyzed to develop two hypothetical multivariate scores concerning the degree to which a patient is affected by the metabolic syndrome. The scores are based on a structural model for low-density cholesterol (LDL) and high-density cholesterol (HDL), triglycerides, uric acid, creatinine, glucose, insulin, systolic blood pressure and waist-to-hip ratio. Age, gender and body mass index were used for adjusting all variables. In segregation analyses of 42 pedigrees without using genotype information, estimations of the heritabilities and environmentally caused variance and covariance components were computed for the individual score values of the two latent factors. The first score shows a heritability of 42%; the environment component disappeared. The score mainly reflects the HDL, LDL and triglyceride levels. The second score shows a heritability of 16% with an environment component of 7%. It includes mainly insulin, uric acid and creatinine. In the search for genetic causes, both scores could be a basis for further phenotypic classification of the metabolic syndrome<br>Dieser Beitrag ist mit Zustimmung des Rechteinhabers aufgrund einer (DFG-geförderten) Allianz- bzw. Nationallizenz frei zugänglich

In a study looking for risk factors of atherosclerosis in families with combined hyperlipidemia and hypertension, clinical and biochemical data of 1,149 persons were analyzed to develop two hypothetical multivariate scores concerning the degree to which a patient is affected by the metabolic syndrome. The scores are based on a structural model for low-density cholesterol (LDL) and high-density cholesterol (HDL), triglycerides, uric acid, creatinine, glucose, insulin, systolic blood pressure and waist-to-hip ratio. Age, gender and body mass index were used for adjusting all variables. In segregation analyses of 42 pedigrees without using genotype information, estimations of the heritabilities and environmentally caused variance and covariance components were computed for the individual score values of the two latent factors. The first score shows a heritability of 42%; the environment component disappeared. The score mainly reflects the HDL, LDL and triglyceride levels. The second score shows a heritability of 16% with an environment component of 7%. It includes mainly insulin, uric acid and creatinine. In the search for genetic causes, both scores could be a basis for further phenotypic classification of the metabolic syndrome.<br>Dieser Beitrag ist mit Zustimmung des Rechteinhabers aufgrund einer (DFG-geförderten) Allianz- bzw. Nationallizenz frei zugänglich.

Il telerilevamento basato su aeromobile a pilotaggio remoto (APR) è stato utilizzato per high-throughput phenotyping (HTP) dei tratti di canapa e miscanto. Il telerilevamento da APR, tramite l’acquisizione di immagini multispettrali ed il calcolo di diversi indici di vegetazione, permette di stimare i tratti delle colture. In questa tesi, i tratti stimati per la canapa sono stati l'indice di area fogliare e il contenuto di clorofilla fogliare, mentre per il miscanto sono stati l'intercettazione della luce, l'altezza della pianta, la biomassa delle foglie verdi, la biomassa epigea e il contenuto di umidità. La stima dei tratti è stata effettuata utilizzando algoritmi di machine learning e l’inversione del modello PROSAIL. L'HTP di canapa e miscanto è stata effettuata applicando il modello additivo generalizzato (GAM) alle serie temporali dei valori dei tratti stimati dai voli APR. Il telerilevamento da APR ha permesso di analizzare le dinamiche dei tratti durante la stagione di crescita. La combinazione di modelli di stima e modellazione GAM, applicata alle serie temporali dei valori dei tratti stimati da più immagini multispettrali acquisite da voli APR, si è rivelata un potente strumento per l'HTP.<br>Unmanned aerial vehicle (UAV) based remote sensing platform was used for high-throughput phenotyping (HTP) of hemp and miscanthus traits. UAV remote sensing, through the acquisition of multispectral images and the calculation of different vegetation indices, is able to estimate the crop traits. In this thesis, the crop traits estimated were leaf area index and leaf chlorophyll content for hemp, while light interception, plant height, green leaf biomass, standing biomass, and moisture content for miscanthus. The estimation of the traits was carried out using machine learning algorithms and the inversion of the PROSAIL model. The HTP of hemp and miscanthus was carried out by applying the generalized additive model (GAM) to the time series of traits values estimated from UAV flights. UAV remote sensing enabled to analyse of the traits' dynamics during the growing season. Combining estimation models and GAM modelling applied to time series of crop trait values estimated from multiple multispectral images of UAV flights proved to be a powerful tool for HTP.

L’estimation de la profondeur consiste à calculer la distance entre différents points de la scène et la caméra. Savoir à quelle distance un objet donné est de la caméra permettrait de comprendre sa représentation spatiale. Les anciennes méthodes ont utilisé des paires d’images stéréo pour extraire la profondeur. Pour avoir une paire d’images stéréo, nous avons besoin d’une paire de caméras calibrées. Cependant, il est plus simple d’avoir une seule image étant donnée qu’aucun calibrage de caméra n’est alors nécessaire. C’est pour cette raison que les méthodes basées sur l’apprentissage sont apparues. Ils estiment la profondeur à partir d’une seule image. Les premières solutions des méthodes basées sur l’apprentissage ont utilisé la vérité terrain de la profondeur durant l’apprentissage. Cette vérité terrain est généralement acquise à partir de capteurs tels que Kinect ou Lidar. L’acquisition de profondeur est coûteuse et difficile, c’est pourquoi des méthodes auto-supervisées se sont apparues naturellement comme une solution. Ces méthodes ont montré de bons résultats pour l’estimation de la profondeur d’une seule image. Dans ce travail, nous proposons d’estimer des cartes de profondeur d’images prises du point de vue des conducteurs de train. Pour ce faire, nous avons proposé d’utiliser les contraintes géométriques et les paramètres standards des rails pour extraire la carte de profondeur à entre les rails, afin de la fournir comme signal de supervision au réseau. Il a été démontré que la carte de profondeur fournie au réseau résout le problème de la profondeur des voies ferrées qui apparaissent généralement comme des objets verticaux devant la caméra. Cela a également amélioré les résultats de l’estimation de la profondeur des séquences des trains. Au cours de ce projet, nous avons d’abord choisi certaines séquences de trains et déterminé leurs distances focales pour calculer la carte de profondeur de la voie ferrée. Nous avons utilisé ce jeu de données et les distances focales calculées pour affiner un modèle existant « Monodepth2 » pré-entrainé précédemment sur le jeu de données Kitti.<br>Depth prediction is the task of computing the distance of different points in the scene from the camera. Knowing how far away a given object is from the camera would make it possible to understand its spatial representation. Early methods have used stereo pairs of images to extract depth. To have a stereo pair of images, we need a calibrated pair of cameras. However, it is simpler to have a single image as no calibration or synchronization is needed. For this reason, learning-based methods, which estimate depth from monocular images, have been introduced. Early solutions of learning-based problems have used ground truth depth for training, usually acquired from sensors such as Kinect or Lidar. Acquiring depth ground truth is expensive and difficult which is why self-supervised methods, which do not acquire such ground truth for fine-tuning, has appeared and have shown promising results for single image depth estimation. In this work, we propose to estimate depth maps for images taken from the train driver viewpoint. To do so, we propose to use geometry constraints and rails standard parameters to extract the depth map inside the rails, to provide it as a supervisory signal to the network. To this end, we first gathered a train sequences dataset and determined their focal lengths to compute the depth map inside the rails. Then we used this dataset and the computed focal lengths to finetune an existing model “Monodepth2” trained previously on the Kitti dataset. We show that the ground truth depth map provided to the network solves the problem of depth of the rail tracks which otherwise appear as standing objects in front of the camera. It also improves the results of depth estimation of train sequences.

During the last half of the 20^th century, the world pork industry has achieved astonishing developments in pig breeding. Now swine farms are larger, ownership more concentrated, and farms have become more industrialized. Artificial insemination (AI) plays an important role in animal breeding increasing utilization of genetically superior sires. Currently boars selected for commercial use as AI sires are evaluated on grow-finish performance and carcass characteristics. The objectives of this study were to (A) estimate genetic correlations between production and semen traits in the boar; average daily gain (ADG), back fat thickness (BF) and muscle depth (MD) as production traits, and total sperm cells (TSC), total concentration (TC), volume collected (SV), number of extended doses (ND), and acceptance rate of ejaculates (AR) as semen traits; (B) to model the variances and covariances of total sperm cells (× 10⁹) over the active lifetime of AI boars; and (C) to compare multiple traits and random regression analyses applied to total sperm cells (TSC). Average heritability estimates were 0.39 for ADG, 0.32 for BF, 0.15 for MD, and repeatability estimates were 0.38 for SV, 0.37 for TSC, 0.09 for TC, 0.39 for ND, and 0.16 for AR. Semen traits showed negative genetic correlations with MD. Genetic correlations would indicate that current selection objectives are having a negative effect on semen traits. Therefore, current AI boar selection practices may be having a detrimental effect on semen production. In random regression analysis for total sperm cells, maximum log likelihood value was observed for sixth, fifth, and seventh order polynomials for fixed, additive genetic and permanent environmental effects, respectively. Best fit as determined by Akaike's Information Criterion was based on a model with sixth, fourth, and seventh order polynomials for fixed, additive genetic and permanent environmental effects, respectively. Best fit as determined by Schwarz Criterion was by fitting fourth, second, and seventh order polynomials for fixed, additive genetic and permanent environmental effects, respectively. Heritability estimates for total sperm cells ranged from 0.27 to 0.61 across age of boar classifications. Heritability for total sperm cells tended to increase with age of boar classification. The cyclic nature of heritability for total sperm cells that was observed over the active lifetime of boars may be due in part to number of observations across seasons limiting our ability to correct for seasonal effects on sperm production. In MTDFREML analysis, heritability estimates of 9, 12, 15, 18, 21, 24, and 27 months of age were, respectively, 0.28, 0.29, 0.26, 0.27, 0.30, 0.79, and 0.41. The results from MTDFREML seemed to be overestimated when compared to random regression. Therefore, random regression methods are the most appropriate to analyze semen traits as they are longitudinal data measured over the boars lifetime.

This thesis addresses the problem of quality degradation in speech produced by parameter-based speech synthesis, within the framework of an articulatory-acoustic forward mapping. I first investigate current problems in speech parameterisation, and point out the fact that conventional parameterisation inaccurately extracts the vocal tract response due to interference from the harmonic structure of voiced speech. To overcome this problem, I introduce a method for estimating filter responses more precisely from periodic signals. The method achieves such estimation in the frequency domain by approximating all the harmonics observed in several frames based on a least squares criterion. It is shown that the proposed method is capable of estimating the response more accurately than widely-used frame-by-frame parameterisation, for simulations using synthetic speech and for an articulatory-acoustic mapping using actual speech. I also deal with the source-filter separation problem and independent control of the voice source characteristic during speech synthesis. I propose a statistical approach to separating out the vocal-tract filter response from the voice source characteristic using a large articulatory database. The approach realises such separation for voiced speech using an iterative approximation procedure under the assumption that the speech production process is a linear system composed of a voice source and a vocal-tract filter, and that each of the components is controlled independently by different sets of factors. Experimental results show that controlling the source characteristic greatly improves the accuracy of the articulatory-acoustic mapping, and that the spectral variation of the source characteristic is evidently influenced by the fundamental frequency or the power of speech. The thesis provides more accurate acoustical approximation of the vocal tract response, which will be beneficial in a wide range of speech technologies, and lays the groundwork in speech science for a new type of corpus-based statistical solution to the source-filter separation problem.

The starting point for this thesis is a concrete problem: to measure, using statistical models, aspects of subjective perceptions and assessments and to understand their dependencies. The objective is to study the statistical properties of some estimators of the parameters of regression models with variables affected by measurement errors. These models are widely used in surveys based on questionnaires developed to detect subjective assessments and perceptions with Likert-type scales. It is a highly debated topic, as many of the relevant aspects in this field are not directly observable and therefore the variables used to estimate them are affected by measurement errors. The models with measurement errors were very thorough in literature. In this work we will developed two of the most used approaches that the authors have with this topic. Obviously, according to the approach chosen, different models were proposed to estimate the relationships between variables affected by measurement error. After exposing the main features of these models, the thesis focuses on providing an original contribution to comparative analysis of the two presented approaches.

Au cours des dernières années, la technologie GNSS a attiré beaucoup d’attention autour du monde et elle a été largement appliquée dans de nombreux domaines. D'autre part, le système d'exploitation ferroviaire avancé a été largement utilisé pour assurer la sécurité, la sûreté et l'efficacité du réseau ferroviaire. L'efficacité de ce système se fonde sur la disponibilité du positionnement fiable du train. L’application de cette technologie au positionnement du train est un domaine de recherche très prometteur. Dans cette thèse, plusieurs algorithmes sont proposés pour le positionnement du train en utilisant des signaux GNSS et un modèle géométrique de voie stocké dans la base de données à bord du train. Premièrement, la distance, vitesse du train sont estimées en utilisant des signaux GNSS et un modèle géométrique ‘idéal’ qui est composé de lignes droites, de courbes de transition et d'arcs de cercle. L’impact du rayon de courbure de la voie sur ces estimations est étudié. Deuxièmement, la distance, vitesse du train sont estimées en utilisant des signaux GNSS et un modèle géométrique ‘non-idéal’ qui est approché par une ligne polygonale avec un certain niveau d'incertitude. L’impact de l’incertitude de la voie sur ces estimations est étudié. Finalement, la distance, vitesse du train sont estimées à l’aide d’intégration des mesures GNSS et une base de données bruitée. L’impact des erreurs de GNSS et de la base de données sur ces estimations est étudié<br>Over these years, GNSS technology has attracted many attentions around world and it has been widely applied in navigation for airplanes, ground vehicles and boats. On the other hand, advanced railway operating systems have been widely used to guarantee the safety and efficiency of the railway network. The efficiency of these systems is based on the availability of reliable train positioning. Hence, applying GNSS technology to the train positioning is a very promising research area, since it has such important benefits as lower initial costs and lower maintenance. In this thesis, several algorithms are proposed for train positioning by using GNSS signals and the railway centerline stored in the onboard computer database. At first, the train travelled distance, speed are estimated by using GNSS signals and an ''ideal'' railway centerline which is composed of straight line segments, transition curves and arcs of circles. The impact of the railroad curvature on these estimations is studied. Secondly, the train travelled distance, speed are estimated by using GNSS signals and a ''non-ideal'' railway centerline which is defined by a polygonal line with some level of uncertainty. The impact of the track geometric model imprecision on these estimations is studied. Finally, the train travelled distance, speed are estimated by integrating the GNSS measurements with a track database. The impact of the GNSS measurements and the track database errors on these estimations is studied

The application of Model Reference Adaptive Control (MRAC) for train dynamic braking is investigated in order to control dynamic braking forces while remaining within the allowable adhesion and coupler forces.  This control method can accurately determine the train braking distance.  One of the critical factors in Positive Train Control (PTC) is accurately estimating train braking distance under different operating conditions.  Accurate estimation of the braking distance will allow trains to be spaced closer together, with reasonable confidence that they will stop without causing a collision.  This study develops a dynamic model of a train consist based on a multibody formulation of railcars, trucks (bogies), and suspensions.   The study includes the derivation of the mathematical model and the results of a numerical study in Matlab.  A three-railcar model is used for performing a parametric study to evaluate how various elements will affect the train stopping distance from an initial speed.  Parameters that can be varied in the model include initial train speed, railcar weight, wheel-rail interface condition, and dynamic braking force.  Other parameters included in the model are aerodynamic drag forces and air brake forces.  <br />An MRAC system is developed to control the amount of current through traction motors under various wheel/rail adhesion conditions while braking.  Minimizing the braking distance of a train requires the dynamic braking forces to be maximized within the available wheel/rail adhesion.  Excessively large dynamic braking can cause wheel lockup that can damage the wheels and rail.  Excessive braking forces can also cause large buff loads at the couplers.  For DC traction motors, an MRAC system is used to control the current supplied to the traction motors.  This motor current is directly proportional to the dynamic braking force.  In addition, the MRAC system is also used to control the train speed by controlling the synchronous speed of the AC traction motors.  The goal of both control systems for DC and AC traction motors is to apply maximum available dynamic braking while avoiding wheel lockup and high coupler forces.  The results of the study indicate that the MRAC system significantly improves braking distance while maintaining better wheel/rail adhesion and coupler dynamics during braking.  Furthermore, according to this study, the braking distance can be accurately estimated when MRAC is used.  The robustness of the MRAC system with respect to different parameters is investigated, and the results show an acceptable robust response behavior.<br /><br>Ph. D.

Positioning systems for trains are traditionally based on track-side infrastructure, implying costs for both installation and maintenance. A reliable on-board system would therefore be attractive. Sufficient reliability for on-board systems is likely going to require a multi-sensor solution. This thesis investigates how measurements from bogie-mounted inertial and magnetic sensors can contribute to such a system. The first part introduces and compares two different methods for estimating the speed. The first one estimates the fundamental frequency of the variations in the magnetic field, and the second one analyses the mechanical vibrations using the accelerometer and gyro, where one mode is due to the wheel irregularities. The second part introduces and evaluates a method for train localization using magnetic signatures. The method is evaluated both as a solution for localization along a given track and at switchways. Overall, the results in both parts show that bogie-mounted inertial and magnetic sensors provide accurate estimates of both speed (within 0.5 m/s typically) and location (3-5 m accuracy typically).

Thesis (S.M. in Transportation)--Massachusetts Institute of Technology, Dept. of Civil and Environmental Engineering, 2010.<br>"September 2010." Cataloged from PDF version of thesis.<br>Includes bibliographical references (p. 114-115).<br>The purpose of this thesis is to assess the feasibility of identifying which trains individual passengers take to get from their origin to destination while travelling in a high frequency urban rail transportation system. If this proves possible, the resulting information will inform capital and operations planning decisions as well as improve the ability to measure the aspects of passenger experience related to travel time and crowding. This thesis will first explore this idea by presenting the design, implementation, and application of a model that attempts to identify the selected train level itineraries through a temporal and spatial matching process. As a result of this process, the model is designed to estimate passenger loads, walk times, and the number of left behind passengers. The thesis will then assess the accuracy of these results by comparing them with figures produced by existing models. The model will be developed and applied in the context of the London Underground, but should also be applicable to other urban public transportation systems. Assessment of the results of this model and consideration of the challenges in the creating the model does not conclusively indicate that identifying the exact train a passenger selects to get from his origin to destination is possible. However, the results do indicate that the model has significant potential, and can be improved in future research. These initial results can serve as indicators on how to improve the model.<br>by Elizabeth Cheriyamadam Paul.<br>S.M.in Transportation

The purpose of this travel cost study is to determine how rider behaviors and site characteristics influence repeat visits for equestrian trail riding in Kentucky. Primary data was collected via a survey developed and administered to trail riders in person and online. The average surveyed trail rider tends to be female, about 46 years old, with some higher education, and an annual household income of $65,000. She makes 11 trips to a specified site per year, 8 of which are daytrips, usually in the fall, and traveling 132 miles round trip. From other information gathered, an index of trail characteristics was developed to identify positive attributes of trails. To account for overdispersion of the number of visits per year, a negative binomial distribution in the estimation was used. The primary variables significant to explaining repeat visits to a site include distance in miles, the index of characteristics, and gender. Given consumer surplus estimates of $800 per equestrian it is recommended that established trails maximize desired characteristics. For new trail development it is recommended that trail characteristics are maximized and that they are built closer to the urban areas of the state since most riders are coming from these areas.

Thesis (MScAgric)--Stellenbosch University, 2008.<br>ENGLISH ABSTRACT: Research conducted on beef cattle in Botswana investigated both growth and reproduction. These studies however, did not specifically determine the influence of the different environmental factors on growth in the Tswana and Composite beef cattle breeds. The establishment of a national beef herd recording and performance testing scheme requires knowledge on the appropriate adjustment methods of field data for the fixed effects such as sex of calf and age of dam. A fair comparison of birth and weaning weights between male and female calves, and calves born from young, mature and old dams will be derived from these adjustment factors. There is no information on adjustment factors for the Tswana and Composite cattle breeds in the country. Genetic parameters for growth traits in these breeds are not known and are needed for the implementation of the performance scheme in Botswana. The Composite breed resulted from a controlled crossbreeding programme using the Simmental, Brahman, Tswana, Tuli and the Bonsmara breeds. The Tswana animals are indigenous to the country and were sourced locally at the beginning of the growth evaluation trial in the two breeds. The objectives of the study were to use data collected from Tswana and Composite cattle breeds to estimate the influence of non-genetic factors on growth traits in the two breeds; to develop adjustment factors for the effects of sex of calf and age of dam; and to estimate genetic parameters (heritabilities and genetic correlations) for future genetic evaluations in both breeds. Data were collected over the period 1988 to 2006. A total of 2 257 records for the Composite breed and 5 923 records for the Tswana breed were available for analyses. Growth characteristics of interest in this study were birth weight (BW), weaning weight (WW), pre-weaning average daily gain (ADG1), 18 months weight (18MW) and post-weaning average daily gain (ADG2). Study 1 indicated that non-genetic effects of breed of calf, sex of calf, month and year of birth, previous parous state, weight of cow at parturition, age of dam, and age of calf at weaning significantly affected BW, WW, 18MW, ADG1 and ADG2 in both breeds. The Composite breed had higher BW, ADG1 and WW whereas the Tswana had higher ADG2 and 18MW. Pre-weaning growth rate increased with an increase in the age of the dam, reaching a peak in mature (5-12 years) cows and declined in cows 13 years and older. Conversely, post-weaning growth rates declined as age of dam advanced but increased in old (13 years and older) dams. Male calves were heavier than female calves for all the growth traits. Birth weight increased as calving season progressed whilst a decrease in WW was observed over the same period. Heifers gave birth to lighter calves when compared to mature multiparous dams. The Composite breed can therefore be considered for weaner production under ranch conditions while the Tswana can be reared under extensive systems due to its adaptability to the environment. Additive correction factors for effects of sex of calf and age of dam on BW and WW were studied separately for the Tswana and Composite in study 2. The least squares means procedure was used to derive age groups and the adjustment factors. The three age groups were young (4 years and below) dams, mature (5-12 years) dams and older (13 years and above) dams. Male calves were heavier than their female counterparts. The sex of calf adjustments for BW and WW were 2.75 and 8.21 kg in the Tswana, and corresponding values for the Composite 2.84 and 10.11 kg, respectively. Birth weight and WW increased as age of dam increased, reached maximum in mature dams and declined in older dams. Age of dam adjustment factors for BW in the 3, 4 and 13+ years age groups for the Tswana were 1.74, 0.96 and 1.87 kg, respectively. The corresponding values for the Composite were 2.28, 0.94 and 2.06 kg, respectively. Age of dam adjustment factors for weaning weight in the Tswana were 10.36 and 5.46 kg for age groups 3-4 and 13+ years, respectively. Adjustment factors for WW in the Composite were 13.84, 3.20 and 9.58 kg for age groups 3, 4 and 13+ years. The differences in adjustment factors obtained between the two breeds emphasize the need to compute and apply these factors within breed. Study 3 involved the estimation of genetic parameters for BW, WW, ADG1, 18MW and ADG2. Single-trait and multi-trait analyses were used in the estimation of (co)variance components by fitting an individual animal model (AM) and the animal maternal model (AMM) for the two breeds. Direct heritabilities for BW, WW, ADG1, 18MW and ADG2 in the Tswana were 0.45, 0.32, 0.37, 0.31 and 0.31, respectively from a single-trait AM analysis. Fitting the AMM resulted in direct heritabilities of 0.31, 0.20 and 0.16 for BW, WW and ADG1, respectively, while the maternal heritabilities were 0.11, 0.15 and 0.21, respectively. For the Composite the direct heritabilities for BW, WW and ADG1 were 0.58, 0.32 and 0.30, respectively with single-trait AM. Partitioning using the AMM resulted in the direct heritabilities for BW, WW and ADG1 of 0.55, 0.17 and 0.14, respectively, while corresponding maternal effects were 0.09, 0.15 and 0.15, respectively. The genetic correlations between direct and maternal effects were positive and ranged from 0.20 to 0.89. When using the multi-trait analysis and fitting the AM, the direct heritabilities for the Tswana were 0.45, 0.37, 0.34, 0.39 and 0.31 for BW, WW, ADG1, 18MW and ADG2, respectively. Genetic correlations between the growth traits ranged from 0.16 to 0.97. Direct (and maternal) heritabilities for BW, WW and ADG1 were 0.31(0.11), 0.19(0.15) and 0.14(0.17), respectively, in the Tswana. Correlations between direct heritabilities for BW, WW and ADG1 ranged from 0.45 to 0.95, while maternal effects ranged from 0.12 to 0.99. The magnitude of the heritabilities indicates an existence of the opportunity to make genetic progress through selection in both breeds. Selection based on WW seems to be the ideal procedure to bring genetic improvement in the Tswana without detrimental long term effects.<br>AFRIKAANSE OPSOMMING: Navorsing wat op die vleisbeesrasse in Botswana gedoen is, het hoofsaaklik op beide groei en reproduksie gehandel. Hierdie studies het egter nie spesifieke gefokus op die bepaling van die invloed wat verskillende omgewingsfaktore op die groei van saamgestelde (d.i. Composite) en die Tswana vleisbeesrasse het nie. Die bepaling van ʼn nasionale vleisbees rekordhouding- en prestasietoetsskema verg kennis van die mees gepaste metode om velddata vir vaste effekte soos geslag van die kalf en ouderdom van die moeder aan te pas. Hierdie aanpassingsmetodes sal lei tot die regverdige vergelyking van geboorte- en speengewigte tussen manlike en vroulike diere, sowel as van kalwers gebore van jong, volwasse of ou moeders. Tans is daar geen inligting oor aanpassingfaktore vir die Tswana en saamgestelde vleisbeesrasse in Botswana bekend nie. Geen genetiese parameters vir groei-eienskappe vir geeneen van die rasse is beskikbaar nie en word benodig vir die implementering van die prestasie skema in Botswana. Die saamgestelde ras is die produk van ʼn beheerde kruisteeltprogram, wat onderskeidelik die Simmental, Brahman, Tswana, Tuli en die Bonsmara beesrasse ingesluit het. Die Tswana ras is inheems aan Botswana en vanaf plaaslike bronne vir die groei evaluasie studie bekom. Die doelwitte van die studie was eerstens die analisering van data wat van beide die Tswana en saamgestelde rasse ingesamel is, om die invloed van nie-genetiese faktore op die groei eienskappe te bepaal om ten einde aanpassingsfaktore vir die effek van geslag van die kalf en ouderdom van die moederdier te ontwikkel. ʼn Tweede doelwit was die bepaling van genetiese parameters (oorerflikhede en genetiese korrelasies) vir die gebruik in toekomstige genetiese evaluering van beide rasse. Data is vanaf 1988 tot 2006 ingesamel. ʼn Totaal van 2 257 waarnemings vir die saamgestelde ras en 5 923 waarnemings vir die Tswana ras is ontleed. Groei eienskappe wat in die studie ondersoek is, het geboortegewig (BW), speengewig (WW), voorspeen gemiddelde daaglikse toename (ADG1), 18-maand gewig (18MW) en naspeense gemiddelde daaglikse toename (ADG2) ingesluit. Studie een het aangedui dat nie-genetiese effekte van die ras van die kalf, die geslag van die kalf, maand en jaar van geboorte, vorige dragtigheidsstatus, koei se gewig met geboorte van kalf, ouderdom van die moederdier en die speenouderdom van die kalf het ʼn betekenisvolle invloed op BW, WW, 18MW, ADG1 en ADG2 van beide rasse gehad. Die saamgestelde ras het hoër waardes vir BW, ADG1 en WW gehad, terwyl die Tswana ras hoër waardes vir ADG2 en 18MW geopenbaar het. Voorspeense groeitempo het toegeneem met ʼn toename in die ouderdom van die moederdier, met ʼn piek in volwasse (d.i. 5-12 jaar ouderdom) moeders en ʼn afname in koeie 13 jaar en ouer. Omgekeerd het naspeen groeitempo afgeneem met ʼn toename in die ouderdom van die moederdier en weer begin toeneem vir ou (d.i. 13 jaar en ouer) koeie. Geboortegewig het toegeneem met die verloop van die kalfseisoen, terwyl ʼn afname in WW vir dieselfde periode aangeteken is. Verse het, wanneer hulle met volwasse koeie vergelyk is, het geboorte aan ligter kalwers gegee. Die saamgestelde ras kan dus oorweeg word vir die produksie van speenkalwers onder kommersiële intensiewe toestande, terwyl die Tswana ras, op grond van sy beter aanpassing by ekstensiewe omstandighede waar die moederlike invloed nie voorkom nie, vir produksie onder ekstensiewe omstandighede gebruik kan word. In studie 2 is die additiewe korreksie faktore vir die invloed van geslag van die kalf en moederouderdom op BW en WW apart vir die twee rasse bestudeer. Die geslag van die kalf x ouderdom van die moederdier interaksie was nie betekenisvol vir enige van die rasse nie. Dus kan geen aanpassing vir die ouderdom van die moeder binne geslagte vir enige van die twee rasse gemaak word nie. Die kleinste kwadraat gemiddeldes metode is gebruik om die ouderdomsgroepe en aanpassingsfaktore te bepaal. Die drie ouderdomsgroepe was jong (d.i. 4 jaar en jonger) koeie, volwasse (d.i. 5-12 jaar ouderdom) en ouer (d.i. 13 jaar en ouer) koeie. Daar is gevind dat manlike kalwers swaarders as hulle vroulike eweknieë is. Die aanpassingswaarde vir die geslag van die kalf vir BW en WW was 2.75 kg en 8.21 kg in die Tswana en 2.84 kg en 10.11kg vir die saamgestelde ras. Geboortegewig en WW het toegeneem met ʼn toename in die ouderdom van die moeder. Dit het ʼn maksimum bereik in volwasse koeie en afgeneem vir koeie ouer as 13 jaar. Die aanpassingsfaktore vir die ouderdom van die moederdier vir BW in die 3, 4 and 13+ jarige ouderdomsgroepe vir die Tswana ras was onderskeidelik 1.74 kg, 0.96 kg en 1.87 kg. Die ooreenstemmende waardes vir die saamgestelde ras was onderskeidelik 2.28 kg, 0.94 kg en 2.06 kg. Aanpassingsfaktore vir WW vir die Tswana ras was 10.36 kg en 5.46 kg vir onderskeidelik die 3-4 jaar en 13+ jaar en ouer ouderdomsgroepe. Aanpassingsfaktore vir WW in die Composite ras was 13.84 kg, 3.20 kg en 9.58 kg vir onderskeidelik die 3 jaar, 4 jaar en 13 jaar en ouer ouderdomsgroepe. Verskille in die onderskeie parameters vir die twee rasse beklemtoon die noodsaaklikheid vir die berekening en toepassing van die onderskeie aanpassingfaktore vir en binne elke ras. Studie 3 het die bepaling van die genetiese parameters vir BW, WW, ADG1, 18MW en ADG2 behels. Enkel- en multivariaat analises is gebruik vir die skatting van die (ko)variansie komponente deur ʼn direkte diermodel (AM) en ʼn dier-maternale model (AMM) vir die twee rasse te pas. Direkte oorerflikhede vir BW, WW, ADG1, 18MW en ADG2 vir die Tswana ras was onderskeidelik 0.45, 0.32, 0.37, 0.31 en 0.31, vir ʼn enkelvariaat AM analise. Die pas van ʼn AMM het direkte oorerflikhede van 0.31, 0.20 en 0.16 vir onderskeidelik BW, WW and ADG1 gegee, terwyl die maternale oorerflikhede onderskeidelik 0.11, 0.15 en 0.21 was. Vir die saamgestelde ras was die direkte oorerflikhede vir BW, WW en ADG1 onderskeidelik 0.58, 0.32 en 0.30 vir die enkelvariaat AM analise. Verdeling (partisie) van die AMM het direkte oorerflikhede vir BW, WW en ADG1 van onderskeidelik 0.55, 0.17 en 0.14 gegee, terwyl die ooreenstemmende maternale effekte onderskeidelik 0.09, 0.15 en 0.15 was. Die genetiese korrelasies tussen die drekte en maternale effekte was positief en tussen 0.20 en 0.89. Met die multivariaat analise en die pas van die AM, is direkte oorerflikhede van 0.45, 0.37, 0.34, 0.39 en 0.31 vir onderskeidelik BW, WW, ADG1, 18MW en ADG2, vir die Tswana ras bereken. Genetiese korrelasies tussen die groei eienskappe het gewissel tussen 0.16 tot 0.97. Direkte (en maternale) oorerflikhede vir BW, WW en ADG1 was onderskeidelik 0.31(0.11), 0.19(0.15) en 0.14(0.17), vir die Tswana ras. Korrelasies tussen die direkte oorerflikhede vir BW, WW en ADG1 het gewissel tussen 0.45 en 0.95, terwyl die maternale effekte tussen 0.12 en 0.99 gewissel het. Die grootte van die oorerflikhede dui op die moontlikheid van genetiese vordering wat deur seleksie in beide rasse gemaak kan word. Seleksie op grond van WW blyk die mees gepaste wyse te wees waarmee genetiese vordering binne die Tswana ras gemaak kan word, sonder enige langtermyn nadelige effekte.

Les progrès récents des réseaux de neurones artificiels (plus connus sous le nom d'apprentissage profond) ont permis d'améliorer l’état de l’art dans plusieurs domaines de la vision par ordinateur. Dans cette thèse, nous étudions des techniques d'apprentissage profond dans le cadre de l’analyse du genre et de l’âge à partir du visage humain. En particulier, deux problèmes complémentaires sont considérés : (1) la prédiction du genre et de l’âge, et (2) la synthèse et l’édition du genre et de l’âge.D’abord, nous effectuons une étude détaillée qui permet d’établir une liste de principes pour la conception et l’apprentissage des réseaux de neurones convolutifs (CNNs) pour la classification du genre et l’estimation de l’âge. Ainsi, nous obtenons les CNNs les plus performants de l’état de l’art. De plus, ces modèles nous ont permis de remporter une compétition internationale sur l’estimation de l’âge apparent. Nos meilleurs CNNs obtiennent une précision moyenne de 98.7% pour la classification du genre et une erreur moyenne de 4.26 ans pour l’estimation de l’âge sur un corpus interne particulièrement difficile.Ensuite, afin d’adresser le problème de la synthèse et de l’édition d’images de visages, nous concevons un modèle nommé GA-cGAN : le premier réseau de neurones génératif adversaire (GAN) qui produit des visages synthétiques réalistes avec le genre et l’âge souhaités. Enfin, nous proposons une nouvelle méthode permettant d’employer GA-cGAN pour le changement du genre et de l’âge tout en préservant l’identité dans les images synthétiques. Cette méthode permet d'améliorer la précision d’un logiciel sur étagère de vérification faciale en présence d’écarts d’âges importants<br>The recent progress in artificial neural networks (rebranded as deep learning) has significantly boosted the state-of-the-art in numerous domains of computer vision. In this PhD study, we explore how deep learning techniques can help in the analysis of gender and age from a human face. In particular, two complementary problem settings are considered: (1) gender/age prediction from given face images, and (2) synthesis and editing of human faces with the required gender/age attributes.Firstly, we conduct a comprehensive study which results in an empirical formulation of a set of principles for optimal design and training of gender recognition and age estimation Convolutional Neural Networks (CNNs). As a result, we obtain the state-of-the-art CNNs for gender/age prediction according to the three most popular benchmarks, and win an international competition on apparent age estimation. On a very challenging internal dataset, our best models reach 98.7% of gender classification accuracy and an average age estimation error of 4.26 years.In order to address the problem of synthesis and editing of human faces, we design and train GA-cGAN, the first Generative Adversarial Network (GAN) which can generate synthetic faces of high visual fidelity within required gender and age categories. Moreover, we propose a novel method which allows employing GA-cGAN for gender swapping and aging/rejuvenation without losing the original identity in synthetic faces. Finally, in order to show the practical interest of the designed face editing method, we apply it to improve the accuracy of an off-the-shelf face verification software in a cross-age evaluation scenario

Cette étude s'intéresse au problème de l'inversion d'un modèle de production de la voix étant donné un enregistrement audio de parole pour obtenir une représentation de le source sonore qui est générée au niveau de la glotte, la source glottique, ainsi qu'un représentation des résonances et anti-résonances créées par les cavités du conduit vocal. Cette séparation des éléments composants la voix donne la possibilité de manipuler indépendamment les caractéristiques de la source et le timbre des résonances. Nous supposons que la source glottique est un signal à phase mixte et que la réponse impulsionnelle du filtre du conduit vocal est un signal à minimum de phase. Puis, considérant ces propriétés, différentes méthodes sont proposées pour estimer les paramètres d'un modèle glottique qui minimisent la phase carrée moyenne du résiduel convolutif d'un spectre de parole observé et de son modèle. Une dernière méthode est décrite où un unique paramètre de forme est solution d'une forme quasi fermée du spectre observé. Ces méthodes sont évaluées et comparées avec des méthodes de l'état de l'art en utilisant des signaux synthétiques et electro-glotto-graphiques. Nous proposons également une procédure d'analyse/synthèse qui estime le filtre du conduit vocal en utilisant un spectre observé et sa source estimée. Des tests de préférences ont été menés et leurs résultats sont présentés dans cette étude pour comparer la procédure décrite et d'autres méthodes existantes.

Master of Agribusiness<br>Department of Agricultural Economics<br>Ted Schroeder<br>Selection of a herd sire has always been of paramount importance given the initial financial investment and their contribution and effect on the genetic make-up of a beef herd. Data was collected from the nation’s longest consecutively run bull test conducted at the University Farm of Oklahoma Panhandle State University (OPSU). The Bull Test and Bull Sale data utilized were collected from 2008-2013. Performance data was collected over a 112 day test period with data collection occurring at 28 day intervals. The top seventy bulls from each year’s test were selected based upon a performance index of ½ ADG and ½ weight per day of age (WDA), and a semen quality and motility score of excellent and sold at auction. Angus bulls were the focus of the study as they represented the vast majority of individuals sold. Three hedonic pricing models were created to try to determine what attributes buyers at the OPSU bull test sale were placing emphasis on. The initial hedonic model contained production data that included BW, ADG, WDA, Julian age, final test weight ultrasound data, and a dummy variable for sale year. The second model utilized all production data and added genetic variables in the form of production EPDs (Calving Ease Direct (CED), BW, Weaning Weight and Yearling Weight) and maternal EPDs (Calving Ease Maternal, Maternal Milk). The third model included the variables from the first and second models with the inclusion of carcass EPDs (Marbling, Ribeye Area (REA) and FAT). Year was significant in all three models however there was less of an effect on price as more variables were included. In model one, the production facts that were of significance were: ADG (P<0.01), BW (P<0.01) and final test weight (P<0.01). In the second model, ADG, BW and final test weight retained their significance at the P<0.01 level. The only production EPD that was significant (P<0.05) was CED. In the third model, years, ADG and BW were still significant (P<0.01). Final test weight (P=0.70) and CED (P = 0.132) had substantial changes. The carcass EPD ribeye area had a P value of 0.057. Producers who are placing bulls on test can utilize the given information to assist with their selection. It cannot go unsaid that while single trait selection can be very detrimental; ADG was significant across all models. The study indicates that performance and growth are of utmost importance to buyers, followed by birth weight consideration.

Thesis (MScAgric )--University of Stellenbosch, 2001.<br>ENGLISH ABSTRACT: The main object of this study was to estimate heritabilities and possible genetic correlations for and between reproductive traits in a composite multi breed beef cattle herd. Reproduction is a complex process with many components. Due to the nature of the data, obtained from the two farms of the Johannesburg Metropolitan Council from 1974 to 1993, only calving date (CD), calving date with a penalty score (CDP), calving success (CS), calving interval (CI), age at first calving (AFC), longevity and stayability at 36, 48, 60, 72 and 84 months were investigated. A GFCAT set of programmes was used and fitted on a sire model to analyse all the categorical traits. Heritabilities and product moment correlations between predicted breeding values for stayability at 36, 48, 60, 72, 84 months, calving success (CS) and longevity, were estimated. The estimated heritabilities on the underlying scale for these traits were 0.06, 0.10, 0.06, 0.03, 0.11, 0.03 and 0.08, respectively. Product moment correlations between breeding values for stayability traits were very low. The highest correlation of 0.22 was obtained between 36 and 48 months. Heritability estimates and correlations between traits appear to be of such a low magnitude that selection for these characteristics would result in limited improvement and indicate that the sire had little influence on his daughter's stayability, longevity and CS. For the analyses oflinear traits (CI, CD, CDP and AFC), a REML procedure fitting a multitrait animal model (using REML VCE 4.2.5 package of Groeneveld, 1998) was used. Heritabilities and geneticcorrelations for and between calving interval (CI), calving date (CD), calving date with a penalty score (CDP) and age at first calving (AFC) were estimated as traits of the dam. The estimated heritabilities for CI, CD, CDP and AFC obtained in this study were 0.01, 0.04, 0.06 and 0.40, respectively with a repeatability of 0.07, 0.12 and 0.13 for CI, CD and CDP, respectively. Genetic correlations between traits obtained varied from low to moderate, except for the high correlations between CD and CDP (0.98), CI and CD (0.75) and CI and CDP (0.79). Heritabilities, genetic correlations and repeatabilities of CD and CDP obtained in this study suggest that CD and CDP are the same traits and that selection for CDP rather than for CD does not have any additional advantage. Due to the additional advantages of CD over CI and the fact that CD is a less biased measurement of the female reproductive complex, CD appears to be of genetic value and should.be considered-as a possible selection criterion to ensure genetic improvement for reproduction in a beef cattle herd.<br>AFRIKAANSE OPSOMMING: Die doeI met hierdie studie was om oorerflikhede vir en moontlike genetiese korrelasies tussen verskillende reproduksiekenmerke in 'n meerras vleisbeeskudde te bepaal. Reproduksie is 'n komplekse proses en bestaan uit veelvuldige komponente. Weens die aard van die data, verkry vanaf die Johannesburg Metropolitaanse Raad se twee plase (vanaf 1974 tot 1993), is slegs kalfdatum (CD), kalf datum met 'n verswaringswaarde (CDP), kalwingsukses (CS), kalfinterval (CI), ouderdom met eerste kalwing (AFC), langlewendheid en volhoubaarheid van die moeders om reproduktief in die kudde tot op onderskeidelik 36, 48, 60, 72 en 84 maande te bly, bestudeer. Kategoriese kenmerke is deur middel van 'n GFCAT stel programme, wat op 'n vadermodel gepas is, geanaliseer. Oorerflikhede vir en die produkmoment korrelasies tussen die voorspelde teelwaardes vir CS, langlewendheid en volhoubaarheid op onderskeidelik 36, 48, 60, 72 en 84 maande, is bereken. Die oorerflikhede vir bogenoemde kenmerke was onderskeidelik 0.03, 0.08, 0.06, 0.10, 0.06, 0.03 en 0.11. Die korrelasie tussen die voorspelde teelwaardes vir die verskillende volhoubaarheidskenmerke was laag. Die hoogste korrelasie, n1. 0.22, is tussen 36 en 84 maande verkry. Die oorerflikhede en korrelasies tussen die kenmerke blyk van so 'n lae omvang te wees dat direkte seleksie vir die kenmerke slegs tot 'n beperkte genetiese verbetering sal lei en dui daarop dat die vader slegs 'n beperkte invloed op CS, volhoubaarheid en die langlewendheid van sy dogters het. Vir die analise van die lineere kenmerke (CI, CD, CDP en AFC), is 'n REML-prosedure gebruik wat op 'n multikenmerk dieremodel gepas is (deur die gebruik van die REML VCE 4.2.5 pakket van Groeneveld, 1998). Oorerflikhede vir en genetiese korrelasies tussen CI, CD, CDP en AFC is bereken as kenmerke van die moeder. In hierdie studie is die beraamde oorerflikhede vir CI, CD, CDP en AFC as onderskeidelik 0.01, 0.04, 0.06 en 0.40 bepaal, met herhaalbaarhede van onderskeidelik 0.07, 0.l2 en 0.13 vir CI, CD en CDP. Genetiese korrelasies tussen die kenmerke het van laag tot matig gevarieer, behalwe vir die hoe korrelasies tussen CD en CDP (0.98), CI en CD (0.75) en CI en CDP (0.79). Die oorerflikhede en herhaalbaarhede vir en genetiese korrelasie tussen CI en CDP verkry in hierdie studie, veronderstel dat CD en CDP in wese dieselfde kenmerk is en dat seleksie vir CDP in plaas van CD geen addisionele voordele inhou nie. Weens die addisionele voordele wat CD inhou, bo die van CI, en die feit dat CD 'n minder sydige bepaling van die vroulike reproduksiekompleks is, blyk CD van genetiese waarde te wees en moet dit as 'n moontlike seleksie kriterium, om genetiese verbetering in 'n vleisbeeskudde te verseker, oorweeg word.

The Piemontese breed is highly specialized for beef production due to double muscling (DM) inherited condition induced by a specific mutation of Myostatin gene (mh) located on Chromosome 2. This phenotype is characterized by greater muscle mass with less fat deposition, reduced weight of the skeleton reduced feed intake, and improved feed conversion. The current breeding goal of the Piemontese population includes traits related to quantitative beef production and maternal calving performance of animals. This thesis is aimed to study the possibility to introduce, indirectly, meat quality traits among the selected traits. This implies that genetic parameters and heritabilities for all meat quality traits need to be estimated. Chapter 3 deals with the development and applications of a calibration curve for Near-Infrared spectroscopy (NIRS) aimed to predict chemical and physical compositions of Piemontese beef cattle. Chapter 4 deals with the estimation of genetic parameters and heritabilities for physical meat quality traits, measured in laboratory, on a dataset of 1,080 Piemontese young bulls using a mixed linear Animal model. Chapter 5 aimed to Predict the fatty acid profiles on a dataset of 733 fresh samples of Longissimus thoracis muscle of Piemontese young bulls using the same regression developed by the calibration curve obtained with near infrared reflectance Spectroscopy (NIRS) technique, and to estimate heritabilities and additive genetic variations for NIRS-predicted traits. Chapter 6 aimed to estimate variance and covariance components and corresponding genetic parameter estimates, and to estimate genetic and phenotypic correlations among meat quality traits in Piemontese beef cattle. (Co)variance components were estimated using a bivariate mixed animal model and was accomplished using Bayesian methodology and the Gibbs sampling integrations. Chapter 7 deals with the estimation of genetic correlations among beef quality traits and performance traits for Piemontese beef cattle, and to evaluate the possibility to include, indirectly, meat quality traits in the selection objective.

Orientador: Lucia Galvão de Albuquerque<br>Banca: Mário Luiz Santana Júnior<br>Banca: Fernando Sebastian Baldi Rey<br>Banca: Ana Fabrícia Braga Magalhães<br>Banca: Roberto Carvalheiro<br>Abstract: Genotype-environment (GxE) interactions could be an important source of variation in reproductive traits with a striking effect on the onset of animal puberty. Thus, the objectives of the present study were: i) to assess the GxE interaction in Nellore sexual precocity indicator traits under different environmental conditions (EC) and ii) to identify, genomic regions and biological pathways associated to Nellore sexual precocity indicator traits and to investigate whether their effects changes according to EC levels. Phenotypic records for age at first calving (AFC), heifer early pregnancy (HP), heifer rebreeding (HR) and scrotal circumference (SC) were collected on 128,994; 85,339; 90,831 and 151,053 animals, respectively. From those, 1800 heifers, 3050 young bulls, and 800 sires were genotyped with BovineHD BeadChip. A reaction norm model was used to estimate the animal's response to environmental conditions changes. To assess the predictive ability the younger scheme and environment-specific scheme were used. For genome-wide scan, the SNP effects for reproductive traits were estimated in three EC levels: Low (EC = -3.0), Medium (EC = 0.0) and High (EC = 3.0) using a linear transformation of the genomic breeding values. The pleiotropic regions associated to reproductive traits (AFC, SC, HP and HR) in three EC levels, were identified using the statistical combination of the single-trait GWAS results and considered significant when -log10(p-valor)>6.0. The inclusion of genomic... (Complete abstract click electronic access below)<br>Resumo: A interação genótipo-ambiente (GxE) pode ser uma importante fonte de variação em características reprodutivas com um efeito notável no início da puberdade animal. Desta forma, os objetivos do presente estudo foram: i) Avaliar a interação GxE em características indicadores de precocidade sexual em animais da raça Nelore em diferentes condições ambientais (EC) e ii) identificar regiões genômicas e vias biológicas associadas a características indicadores de precocidade sexual e verificar se seus efeitos mudam de acordo com os níveis de EC. Informações fenotípicas para idade ao primeiro parto (AFC), ocorrência de prenhez precoce (HP), reconcepção de novilhas (HR) e perímetro escrotal (SC), foram coletados em 128.994, 85.339, 90.831 e 151.053 animais, respectivamente. Destes, 1800 novilhas, 3050 touros jovens e 800 touros foram genotipados com BovineHD BeadChip. Um modelo de norma de reação foi usado para estimar a resposta do animal às mudanças nas condições ambientais. Para avaliar a capacidade preditiva, foram utilizados os esquemas de validação em animais jovens e em ambiente específico. Para varredura genômica ampla os efeitos dos marcadores SNP para as características reprodutivas foram estimados em três níveis de EC Baixo (EC = -3.0), Médio (EC = 0.0) e Alto (EC = 3.0) usando uma transformação linear dos valores genômicos genéticos. As regiões pleiotrópicas associadas com características reprodutivas (AFC, SC, HP e HR) em três EC foram identificadas utilizando a combinação ... (Resumo completo, clicar acesso eletrônico abaixo)<br>Doutor

Thesis (MScAgric)--Stellenbosch University, 2012.<br>ENGLISH ABSTRACT: Profitable milk production and genetic improvement in dairy herds are largely dependant on fertile cows capable of calving down on an annual basis. Several studies indicate declines in the reproductive performance of Holstein cows over the last 30 years. Calving interval (CI) and services per conception (SPC) are being used by dairy farmers as indicators of the reproductive performance of dairy cows. However, using these traits as cow fertility indicators is problematic as CI is dependent on subsequent calving dates while SPC is strongly linked to inseminator proficiency. The aim of the study is therefore, firstly, to describe alternative fertility traits derived from insemination and calving dates and pregnancy check results. The effects of some non-genetic factors on these traits are discussed. Means±sd for interval traits from calving to first insemination (CFS) and the interval from calving to conception (DO) were respectively 77±30 and 134±74 days while the number of services per conception (SPC) averaged 2.55±1.79. The percentage of first servics occurring within 80 days post-partum (FS80d) and the proportion of cows being confirmed pregnant within 100 (PD100d) and 200 days post-partum (PD200d) were 0.64±0.48, 0.36±0.48 and 0.71±0.45, respectively. Although fertility traits were affected significantly by lactation number, calving year and month, herds (managers) had the largest effect. Fertility is a complex trait, however, the challenge is finding traits that best describe this trait. Genetic parameters for these traits could give an indication of the response to selection in dairy herds. In the second part of this study, genetic parameters and correlations between fertility traits, sourced from standard reproduction management data bases, are analysed for Holstein cows using bivariate linearlinear and linear-threshold animal models. Insemination events (n = 69 181) from 26 645 lactations of 9 046 Holstein cows from 14 herds, calving down during the period from 1991 to 2007, were available. The outcome of each AI event was known. Insemination records were linked to the calving date of each cow, lactation number as well as dam and sire identification. Fertility traits indicating the ability of cows to show heat early in the breeding period, and to become pregnant, were derived. Data were analysed using bivariate linear-linear and linear-threshold animal models with fixed effects being herd (14 levels), year (17 levels), season (4 levels) and lactation number (6 levels). The model included the random effects of animal and animal permanent environment (PE). Heritability estimates ranged from 0.04±0.01 to 0.10±0.02 for FS80d, from 0.07±0.01 to 0.08±0.02 for PD100d and from 0.06±0.04 to 0.08±0.02 for PD200d depending on the two-trait combination. Although heritability estimates of most fertility traits were below 0.10, they were in close agreement with results published by other researchers using linear models. Genetic correlations between different fertility parameters analyzed in this study indicated that it is unlikely that a single characteristic would serve well for selection purposes; instead, combining different traits could be used in selection programmes to improve fertility. Further research in constructing an optimal fertility index is warranted. In the third part of this study, genetic parameters for South African Holstein cows for fertility and production traits were estimated from 2415 lactation records. Two-trait analysis of fertility and milk yield was investigated as a method to estimate fertility breeding values when culling, or selection based on milk yield in early lactation, determines presence or absence of fertility observations in later lactations. Fertility traits were days from calving to first service (CFS), days from calving to conception (DO), percentage cows inseminated within 80d post-partum (FS80d), number of service per conception (SPC), and the binary traits percentage of cows pregnant within 100d and 200d postpartum (PD100d, Pd200d). Milk production traits were 300 day milk, fat and protein yield. For fertility traits, range of estimates of heritability (h2) was 0.006 to 0.08 for linear traits and 0.05 to 0.12 for binary traits. The range for permanent environmental variance (c2) was 0.016 to 0.032. In this study genetic correlations of fertility with milk production traits were unfavourable ranging between −0.93 to 0.76.<br>AFRIKAANSE OPSOMMING: Winsgewende melkproduksie en genetiese verbetering in melkkuddes hang grootliks af van vrugbare koeie wat op ‘n jaarlikse basis kalf. Verskeie studies toon dat die reproduksievermoë van Holsteinkoeie oor die afgelope 30 jaar afgeneem het. Melkboere gebruik tussenkalfperiode (TKP) en aantal inseminasies per konsepsie (KIPK) as aanduidings van die reproduksievermoë van melkkoeie. Dit is egter moeilik om hierdie eienskappe as vrugbaarheidseienskappe vir melkkoeie te gebruik omdat TKP afhanklik is van opeenvolgende kalfdatums terwyl KIPK baie sterk gekoppel is aan die inseminasievermoë van die insemineerders. In die eerste gedeelte van die studie word alternatiewe vrugbaarheidseienskappe, wat afgelei is van inseminasie- en kalfdatums en die uitslag van dragtigheidsondersoeke, beskryf. Die invloed van ‘n aantal nie-genetiese faktore op dié eienskappe word ook bespreek. Gemiddeldes±standaard afwykings vir die periode vanaf kalwing tot eerste inseminasie (CFS), die periode van kalwing tot konsepsie (DO) was 77±30 en 134±74 dae onderskeidelik, terwyl die aantal inseminasies per konsepsie (SPC) 2.55±1.79 was. Die persentasie eerste inseminasies wat binne 80 dae na kalwing (FS80d), en die persentasie koeie wat dragtig bevestig is binne 100 (PD100d) en 200 dae na kalwing (PD200d) was 0.64±0.48, 0.36±0.48 en 0.71±0.45, onderskeidelik. Hoewel vrugbaarheidseienskappe betekenisvol beïnvloed is deur laktasienommer, jaar en seisoen van kalwing, het kuddes (bestuurders) die grootste effek op eienskappe gehad. Vrugbaarheid is ‘n ingewikkelde eienskap en die uitdaging is om eienskappe te vind wat dit die beste beskryf. Genetiese parameters vir eienskappe wat oorweeg word sal ‘n aanduiding gee van die seleksieresponse in melkkuddes. In die tweede gedeelte van die studie is genetiese parameters van vrugbaarheidseienskappe en korrelasies tussen dié eienskappe beraam. Eienskappe is bekom vanaf reproduksiebestuursprogramme wat in melkkuddes gebruik word. Al die inseminasierekords (n = 69 181) van 26 645 laktasies van 9 046 Holsteinkoeie van 14 melkkuddes wat tussen 1991 en 2007 gekalf het, was beskikbaar. Die uitslag van elke inseminasie was bekend. Inseminasierekords is met die kalfdatum, laktasienommer, identifkasienommers van die moeder en vader van elke koei, gekoppel. Vrugbaarheidseienskappe wat die vermoë van koeie aandui om vroeg na kalwing op hitte te kom en beset te raak, is verkry. Die data is ontleed deur gebruik te maak van twee-eienskap liniêreliniêr- en liniêr-drempel-diere modelle met vaste effekte kudde (14 vlakke), jaar (17 vlakke), siesoen (4 vlakke) en laktasienommer (6 vlakke). Die modelle het die ewekansige effekte van dier- en dierpermananente omgewingseffkte (PE) ingesluit. Genetiese, diere PE en residuele korrelasies is vervolgens beraam. Oorerflikhede varieer vanaf 0.04±0.01 tot 0.10±0.02 vir FS80d, vanaf 0.07±0.01 tot 0.08±0.02 vir PD100d en vanaf 0.06±0.04 tot 0.08±0.02 vir PD200d, afhangende van die tweeeienskap kombinasie. Ten spyte daarvan dat die oorerflikhede van die meeste vrugbaarseienskappe laer as 0.10 is, is die resultate in ooreenstemming met ander navorsers wat liniëre modelle gebruik het. Genetiese korrelasies tussen verskillende vrugbaarheidseienskappe toon dat daar nie enkel beste eienskap is wat vir seleksiedoeleindes gebruik kan word nie. Dit sou waarskynlik beter wees om verskillende eienskappe te kombineer om die vrugbaarheid in melkkoeie te verbeter. Verdere navorsing is nodig om ‘n optimale vrugbaarheidseienskap te ontwikkel. Dit is belangrik dat die verband tussen reproduksie en melkproduksie vir melkkoeie bepaal word. In die derde gedeelte van die studie is genetiese parameters vir vrugbaarheid- en melkproduksieeienskappe vir Suid Afrikaanse Holsteinkoeie beraam. ‘n Totaal van 2415 laktasierekords was beskikbaar. Vrugbaarheid en melkproduksie is volgens ‘n twee-eienskap analise ontleed as ‘n metode om teelwaardes vir vrugbaarheid te bepaal in gevalle waar die uitskot of seleksie gebaseer op melkproduksie in vroeglaktasie die teenwoordigheid of afwesigheid van vrugbaarheidseienskappe in latere laktasies bepaal. Vrugbaarheidseienskappe was die periode (aantal dae) tussen kalfdatum en eerste inseminasie (CFS), die aantal dae van kalf tot konsepsie (DO), die persentasie koeie wat by 80 dae na kalf vir die eerste keer geïnsemineer was (FS80d), die aantal inseminasies per konsepsie (SPC), en binêre eienskappe van die persentasie koeie wat by 100 dae en 200 dae na kalf beset was (PD100d en PD200d). Melkproduksie-eienskappe was 300-dae melk., vet- en proteïnproduksie. Vir vrugbaarheidseienskappe het die oorerflikheidswaardes (h2) vanaf 0.006 tot 0.08 vir liniêre eienskappe gevarieer en tussen 0.05 tot 0.12 vir binêre eienskappe. Die permanente omgewingsvariansie (c2) het tussen 0.016 tot 0.032 gevarieer. In hierdie studie was die genetiese korrelasies tussen vrugbaarheidseienskappe en melkproduksie-eienskappe ongunstig en het dit tussen −0.93 tot 0.76 gevarieer.

Non-additive effects may affect the evolution of populations by lowering the heritability of the traits that they affect, thus causing inbreeding depression within populations and playing a role in the conversion of non-additive into additive variation during bottlenecks and in the evolution and maintenance of negative genetic correlations between traits (Crnokrak and Roff 1995, Wolf et al. 2000). Furthermore, dominance variance should be present to a greater degree in traits closer to fitness (Crnokrak and Roff 1995). This study uses diallel cross analyses of inbred lines of the sand cricket Gryllus firmus, to examine the sources of variation in weight at age, adult size, development time, fecundity and flight muscle weight and in particular the ratios of additive, dominance and maternal variance to total variance. We also examine the genetic relationships between the traits.<br>Our study also examines the presence of maternal effects in growth traits and adult size in the sand cricket Gryllus firmus using diallel cross analyses of inbred lines.

Les nuages de glace de type cirrus sont reconnus comme ayant un impact radiatif important mais encore mal déterminé sur le système Terre-atmosphère. Récemment, la constellation de satellites A-Train a efficacement contribué à leur étude, grâce un panel d’instruments en parfaite synergie les uns avec les autres. Dans cette thèse, nous proposons d’utiliser l’instrumentation de l’A-Train afin d’étudier les propriétés optiques et radiatives des cirrus. Dans un premier temps, des données issues de deux campagnes aéroportées sont utilisées, afin d’effectuer une validation des mesures du radiomètre infrarouge spatial IIR. Nous montrons de très bonnes similitudes entre les mesures radiométriques aéroportées et spatiales, ce qui permet de conclure à la validation de ces dernières. Une seconde étude présente un algorithme développé dans le but de restituer l’épaisseur optique des cirrus et la dimension effective des cristaux qui les composent, à partir des mesures de IIR. Nous montrons que ses résultats sont en accord avec des mesures in situ et des produits opérationnels, mais remarquons cependant que leur qualité pourrait être améliorée dans l’hypothèse d’une meilleure connaissance des propriétés de nuages d’eau liquide sous-jacents. Une troisième étude propose donc une amélioration de cet algorithme, permettant de restituer simultanément les propriétés d’une couche de nuage de glace et de deux couches de nuage d’eau liquide. Cet algorithme ’multi-couches’ est appliqué sur un grand nombre de cas, de manière à juger efficacement de la qualité de ses résultats. Des comparaisons avec divers produits opérationnels montrent une bonne cohérence de nos restitutions<br>Cirrus are cloud types that are recognized to have a strong but still poorly understood impact on the Earth-atmosphere radiation balance. Recently, the A-Train satellite constellation has efficiently contributed to the study of these clouds, due to a multitude of instruments in perfect synergism. In this PhD research, several A-Train instruments have been used to study the optical and radiative properties of cirrus. Firstly, with the help of the data from two airborne campaigns, we have performed a validation of the measurements from the Infrared Imaging Radiometer (IIR) onboard CALIPSO. We have observed strong similarities between the airborne and space borne radiometric measurements, which allows validating the latter. A second study presents an algorithm developed in order to perform retrievals of the optical thickness of cirrus and the effective size of their ice crystals using IIR measurements. We demonstrate that these retrievals are perfectly coherent with in situ measurements and operational products of IIR. It is nevertheless observed that better constraints on the properties of liquid water clouds underneath cirrus layers could significantly improve the retrievals. Therefore, a third study presents a modification of this algorithm, allowing us to simultaneously retrieve the properties of one cirrus layer and two layers of liquid water clouds underneath. This ‘multi-layer’ algorithm is applied to a large amount of cases in order to assess its effectiveness. Comparisons with several operational products indicate a good coherence of our retrievals

Railway signaling is of paramount importance to ensure traffic management andsafety on the rail network. The main lines are divided into sections called ‘blocks’,which are governed by a fixed signal installation. To prevent trains from colliding,each block allows one train at once. In France (and most European countries),train detection is performed by an electrical device called track circuit that consistsof a transmitter and a receiver installed at the track-side, and connected via therails. In the absence of a train, an electrical signal flows from the transmitter tothe receiver through the rails. As a train enters a track circuit, its axles shuntthe rails, provoking a short circuit (also called ‘shunt’): the signal transmitted tothe receiver drops. The detection of that signal drop results in the detection of atrain. This method rarely fails throughout the network, but there can be criticalcases where it may be inefficient. In this Master’s Thesis, new detection criteriaproposed in previous studies have been tested on signals measured in poor shuntingconditions. Three approaches have been tested: one in the time domain and two inthe frequency domain. The time approach compares the short-term and long-termstatistics of the received signals. The observation of a change in the spectra of thereceived signals around the 3rd order harmonic (3OH) has led to the implementationof two frequency criteria: the estimation of the band power around the 3OH andthe detection of the 3OH peaks. The obtained results show that better detection isachieved when the new criteria and the existing one are combined.<br>Tågsignalsystem är väsentliga för att garantera trafikstyrning och säkerhet i tågnätet.Spåren är indelade i sektioner, s.k. block, som övervakas med fasta signalinstallationer.För att hindra tåg från att krocka, tillåts bara ett tåg i taget per block. IFrankrike (och de flesta andra europeiska länder), detekteras tågen med en elektriskspårkrets som består av en sändare och en mottagare som är installerad bredvidspåret och ansluten till rälsen. När inget tåg finns på spåret, flyter en elektrisk signalfrån sändaren till mottagaren via spåret. När ett tåg anländer, kortsluts kretsenav hjulaxeln och signalen försvinner från mottagaren. Minskningen i signalstyrkaanvänds för att detektera tåget. Denna metod sällan misslyckas i tågnätet, men iovanliga fall kan det uppstå farliga situationer. I detta examensarbete utvärderasnya detektionsmetoder, som har föreslagits i tidigare studier, på signaler som haruppmätts under förhållanden med dålig kontakt mellan hjul och spår. Tre olika metoderhar testats, en i tidsdomänen och två i frekvensdomänen. Tidsdomänsmetodenjämför kortvarig och långvarig statistik för den mottagna signalen. I spektrum förden mottagna signalen, har man observerat en förändring runt den tredje övertonen,samt detektering av frekvenstoppar vid tredje övertonen. De erhållna resultatenvisar på förbättrad detektering när de nya och existerande kriterierna kombineras.

This master thesis is a part of the project which is conducted by Prof.Gianluca Palli in cooperation with a private company in Laboratory of Automation and Robotics (LAR) of the University of Bologna. The project aims to control the given washing machines in a laundromat using mobile based manipulator and visualize all information in a compared manner for the user. The first goal of this thesis is to interpret data on the user interface of the washing machine using Deep Learning and compare it with actual data acquired from the machine itself. The focus lies on visualizing these two type of information directly to the user with the result of this comparison on GUI application. The second goal is to detect the washing machine and its pieces in certain area using convolutional neural networks. Deep learning has absolutely dominated computer vision over the last few years, achieving top score on many task in this field, with neural networks repeatedly push- ing the frontier of visual recognition. After introduction basic information about Deep Learning and particularly MobileNet architectures, Image Processing and Computer Vision concepts are presented. These concepts such as Homography matrix estimation, are used to automatically localize the region of interests of the user interface on the scene frame without using any external markers and recognize the washing program, option and functionalities as well with help of OpenCV library. Next, training and validation procedures of MobileNetV2 for both visual recognition and object detection tasks are described. The results are shown after finetuning with sample data using TF-Slim image recognition library and training the network for object detection using Tensorflow’s Object Detection API. Finally, Loss function on training and validation are indicated and the results on the scene frame visually are shown in a systematical way and further possible application are briefly discussed.

Le traitement des données massives, communément connu sous l’appellation “Big Data”, constitue l’un des principaux défis scientifiques de la communauté STIC.Plusieurs domaines, à savoir économique, industriel ou scientifique, produisent des données hétérogènes acquises selon des protocoles technologiques multi-modales. Traiter indépendamment chaque ensemble de données mesurées est clairement une approche réductrice et insatisfaisante. En faisant cela, des “relations cachées” ou des inter-corrélations entre les données peuvent être totalement ignorées.Les représentations tensorielles ont reçu une attention particulière dans ce sens en raison de leur capacité à extraire de données hétérogènes et volumineuses une information physiquement interprétable confinée à un sous-espace de dimension réduite. Dans ce cas, les données peuvent être organisées selon un tableau à D dimensions, aussi appelé tenseur d’ordre D.Dans ce contexte, le but de ce travail et que certaines propriétés soient présentes : (i) avoir des algorithmes de factorisation stables (ne souffrant pas de probème de convergence), (ii) avoir un faible coût de stockage (c’est-à-dire que le nombre de paramètres libres doit être linéaire en D), et (iii) avoir un formalisme sous forme de graphe permettant une visualisation mentale simple mais rigoureuse des décompositions tensorielles de tenseurs d’ordre élevé, soit pour D &gt; 3.Par conséquent, nous nous appuyons sur la décomposition en train de tenseurs (TT) pour élaborer de nouveaux algorithmes de factorisation TT, et des nouvelles équivalences en termes de modélisation tensorielle, permettant une nouvelle stratégie de réduction de dimensionnalité et d'optimisation de critère des moindres carrés couplés pour l'estimation des paramètres d'intérêts nommé JIRAFE.Ces travaux d'ordre méthodologique ont eu des applications dans le contexte de l'analyse spectrale multidimensionelle et des systèmes de télécommunications à relais<br>Massive and heterogeneous data processing and analysis have been clearly identified by the scientific community as key problems in several application areas. It was popularized under the generic terms of "data science" or "big data". Processing large volumes of data, extracting their hidden patterns, while preforming prediction and inference tasks has become crucial in economy, industry and science.Treating independently each set of measured data is clearly a reductiveapproach. By doing that, "hidden relationships" or inter-correlations between thedatasets may be totally missed. Tensor decompositions have received a particular attention recently due to their capability to handle a variety of mining tasks applied to massive datasets, being a pertinent framework taking into account the heterogeneity and multi-modality of the data. In this case, data can be arranged as a D-dimensional array, also referred to as a D-order tensor.In this context, the purpose of this work is that the following properties are present: (i) having a stable factorization algorithms (not suffering from convergence problems), (ii) having a low storage cost (i.e., the number of free parameters must be linear in D), and (iii) having a formalism in the form of a graph allowing a simple but rigorous mental visualization of tensor decompositions of tensors of high order, i.e., for D&gt; 3.Therefore, we rely on the tensor train decomposition (TT) to develop new TT factorization algorithms, and new equivalences in terms of tensor modeling, allowing a new strategy of dimensionality reduction and criterion optimization of coupled least squares for the estimation of parameters named JIRAFE.This methodological work has had applications in the context of multidimensional spectral analysis and relay telecommunications systems

Esta dissertação tem como objetivo estudar um método para estimar a velocidade de veículos ferroviários usando processamento de vídeo. O sistema proposto consiste na utilização de câmeras ao longo das vias férreas que permitam não só o monitoramento do tráfego ferroviário, mas cujo vídeo captado possa ser utilizado para a obtenção de estimativas para a velocidade instantânea dos trens que por ela trafegam. Tal sistema seria utilizado independentemente dos sistemas de controle já utilizados pela operadora do sistema ferroviário, permitindo que os controladores possam ter uma segunda análise no caso de falha da primeira, assim como um modelo que permita avaliar a velocidade instantânea do veículo ferroviário ao longo do percurso. Os algoritmos de rastreamento empregados para esse fim abordaram diferentes métodos. Confrontaram-se os resultados obtidos com os algoritmos propostos com os dados empíricos de forma a determinar aquele com melhor resposta dada as características do sistema. O algoritmo que apresentou os melhores resultados emprega um único bloco de referência para todos os quadros comparados. A métrica de similaridade responsável por determinar quais blocos são mais ou menos similares dentro do universo de busca estipulado é a soma de diferenças absolutas (SAD, Sum of Absolute Differences). O tempo de processamento requerido por cada um dos métodos de rastreamento estudados também foi considerado nas avaliações de resultados apresentadas. Uma comparação realizada entre as velocidades coletadas e aquelas informadas pelo sistema de controle mostraram que os resultados obtidos com o sistema atual, utilizando a sinalização apenas por circuito de via apresenta resultados pouco confiáveis com erros bastante significativos. Os resultados obtidos com o sistema proposto apresentaram erros menores quando comparados àqueles obtidos pelo sistema vigente, apresentando-se assim como uma solução viável e de baixo custo quando comparada às técnicas atualmente empregadas para a medida de velocidade de trens.<br>This dissertation aims to study a method of estimating the speed of railway vehicles using video processing. The proposed system uses cameras disposed alongside railways to monitor traffic as well as estimate train speeds. Such a system would allow the machinist to have a backup control system, regardless of the control systems already used by the operator of the rail system. In this dissertation, tracking algorithms with different methods are discussed. By comparing results obtained with the proposed algorithm and results from empirical data, it was possible to determine the one that best provided the desired results given the characteristics of the system. The algorithm which presented the best results was the one that employs a single block for all reference frames compared. The metric of similarity used for determining which blocks are most similar within the searched universe is the sum of absolute differences (SAD). The processing time required by each of the screening methods studied was also considered in assessing the results presented. A comparison between the rates collected and those informed by the control system showed that the results obtained with the current system, which only uses the signal track circuit, provides unreliable results with very significant errors. The results obtained with the proposed system had errors smaller than those obtained with the current system, suggesting the proposed system is viable and cost-effective compared to currently employed techniques.

Computer based simulation models have often been used to study the multimodal freight transportation system. But these studies have not been able to dynamically couple the various modes into one model; therefore, they are limited in their ability to inform on dynamic system level interactions. This research thesis is motivated by the need to dynamically couple the multimodal freight transportation system to operate at multiple spatial and temporal scales. It is part of a larger research program to develop a systems modeling framework applicable to freight transportation. This larger research program attempts to dynamically couple railroad, seaport, and highway freight transportation models. The focus of this thesis is the development of the coupled railroad and seaport models. A separate volume (Wall 2010) on the development of the highway model has been completed. The model railroad and seaport was developed using Arena® simulation software and it comprises of the Ports of Savannah, GA, Charleston, NC, Jacksonville, FL, their adjacent CSX rail terminal, and connecting CSX railroads in the southeastern U.S. However, only the simulation outputs for the Port of Savannah are discussed in this paper. It should be mentioned that the modeled port layout is only conceptual; therefore, any inferences drawn from the model's outputs do not represent actual port performance. The model was run for 26 continuous simulation days, generating 141 containership calls, 147 highway truck deliveries of containers, 900 trains, and a throughput of 28,738 containers at the Port of Savannah, GA. An analysis of each train's trajectory from origin to destination shows that trains spend between 24 - 67 percent of their travel time idle on the tracks waiting for permission to move. Train parking demand analysis on the adjacent shunting area at the multimodal terminal seems to indicate that there aren't enough containers coming from the port because the demand is due to only trains waiting to load. The simulation also shows that on average it takes containerships calling at the Port of Savannah about 3.2 days to find an available dock to berth and unload containers. The observed mean turnaround time for containerships was 4.5 days. This experiment also shows that container residence time within the port and adjacent multimodal rail terminal varies widely. Residence times within the port range from about 0.2 hours to 9 hours with a mean of 1 hour. The average residence time inside the rail terminal is about 20 minutes but observations varied from as little as 2 minutes to a high of 2.5 hours. In addition, about 85 percent of container residence time in the port is spent idle. This research thesis demonstrates that it is possible to dynamically couple the different sub-models of the multimodal freight transportation system. However, there are challenges that need to be addressed by future research. The principal challenge is the development of a more efficient train movement algorithm that can incorporate the actual Direct Traffic Control (DTC) and / or Automatic Block Signal (ABS) track segmentation. Such an algorithm would likely improve the capacity estimates of the railroad network. In addition, future research should seek to reduce the high computational cost imposed by a discrete process modeling methodology and the adoption of single container resolution level for terminal operations. A methodology combining both discrete and continuous process modeling as proposed in this study could lessen computational costs and lower computer system requirements at a cost of some of the feedback capabilities of the model This tradeoff must be carefully examined.

Tato disertační práce je zaměřena na analýzu řečového signálu za učelem detekce nemocí ovlivňujících strukturu hlasových orgánů, obzvláště těch, které mění strukturální character hlasivek. Poskytnut je přehled současných technik. Dále jsou popsány zdroje použitých nahrávek pro zdravé a nemocné mlučí. Hlavním učelem této disertační práce je popsat vypočetní postup k odhadu parametrů modelu hlasového zdroje, které umožní následnou detekci a klasifikaci nemocí hlasivek. Poskytujeme detailní popis analýzy řečových signálů, které mohou být odvozeny z parametrických modelů hlasivek.

Glucosinolates have been implicated as a mediator of the cancer-protective properties of cruciferous vegetables. Enzymatic hydrolysis of glucosinolates by plant or microbial myrosinase yields a range of metabolites including beneficial isothiocyanates. Little is known about the fate of glucosinolates after their ingestion. Using urinary end-products of metabolism as markers, measurement of the production of isothiocyanates in the digestive tract of monogastric animals has been achieved. Initially, a range of isothiocyanates were administered to rats and their excretion as mercapturic acids was quantified. Relative recovery of different isothiocyanates was found to be consistent and predictable, allowing the use of artificial isothiocyanates as recovery standards in subsequent experiments. Subsequently, the relative influence of plant and bacterial myrosinase on isothiocyanate production was quantified in rats. A proportion of 0.80 (s. e. m. 0.076) of benzyl glucosinolate was hydrolysed to isothiocyanate by plant myrosinase. In the presence of both plant and microbial activity,, the proportion of benzyl isothiocyanate release was significantly decreased (0.50 s. e. m. 0.046, p<0.01) suggesting microbial breakdown of isothiocyanates. The approach, adapted for use with human subjects showed that the proportions of allyl isothiocyanate measured after ingestion of raw and cooked cabbage were 0.37 (s. e. m. 0.045) and 0.53 (s. e. m. 0.134) respectively in healthy male volunteers. A further experiment with rats established that isothiocyanate uptake in the distal digestive tract was significantly less than in the proximal intestine (0.12 s. e. m. 0.017 and 0.48 s. e. m. 0.029 respectively), suggesting a potential underestimation of isothiocyanate release in the distal digestive tract when using urinary markers. Finally, enhancement of bacterial fermentation by addition of inulin to the diet had little influence on isothiocyanate production in the gut. The findings suggested that the formation of the cancer-protective isothiocyanates was significant, in vivo, thereby strengthening the evidence for a beneficial effect of cruciferous vegetables for health. The newly developed method opens up possibilities of concurrently exploring the digestive fate of isothiocyanates and the toxicity of carcinogenic compounds.

L’objectif principal de cette thèse est de développer une méthodologie qui permette de définir des sollicitations simplifiées représentatives des sollicitations réelles rencontrées en exploitation, c’est-à-dire équivalentes en sévérité vis-à-vis du phénomène d’amorçage de fissure par fatigue mécanique. Cette méthodologie doit être adaptée aux problèmes multi-paramètres car les sollicitations subies par les roues ferroviaires sont multidimensionnelles (chargement vertical et latéral) et indépendantes. Enfin, la thèse vise aussi à proposer une approche fiabiliste globale du problème de fatigue des roues, qui pourra être une extension de la méthode Contrainte-Résistance aux cas des sollicitations multi-entrée. Un critère de fatigue pour l’acier des roues est identifié à partir des nouveaux essais<br>The main objective of this PHD thesis is to develop a method for the definition of simplified equivalent loads representative of real loads (the severity is equivalent from fatigue phenomenon point of view). This method must be adapted to multi-input problems because loads applied on wheels are multi-dimensional (vertical and lateral loads) and independent. Finally, the thesis also aims to provide a comprehensive approach to fatigue-reliability problem of the wheels, which can be an extension of the stress-strength method for multi-input loads. A fatigue criterion for the railway will is presented and identified using a new test campaign

Le progrès génétique est l'un des principaux leviers utilisés pour accroître la production alimentaire et nourrir la population humaine croissante dans un contexte de changement global. La sélection ou la création du cultivar optimal pour un endroit donné est très difficile compte tenu de la très grande variabilité spatiale et temporelle des conditions environnementales. Le phénotypage au champ, c'est-à-dire le suivi quantitatif des variables d'état des cultures et du fonctionnement du couvert, a été reconnu comme le goulot d'étranglement pour accélérer le progrès génétique. Dans cette thèse pluridisciplinaire, des méthodes statistiques et de traitement d’images sont développées afin d’estimer différents traits structuraux de cultures de blé pour application à l’amélioration variétale. Cette thèse a été entreprise au moment où la technologie progresse très rapidement, à la fois sur les aspects matériels et logiciels : accessibilité aux plates-formes de drones et de véhicules sans pilote, diminution du coût des unités de traitement graphique (GPU) explosion du développement des algorithmes d'apprentissage profond. Cette thèse s’articule en cinq chapitres : Le premier chapitre introduit les motivations de l’étude ainsi que les besoins actuel en matière de phénotypage haut débit. Un état de l’art sur le phénotypage est aussi présenté en attirant l’attention sur les méthodes de traitements d’images et de réseaux de neurones convolutifs. Le deuxième chapitre présente le développement de méthodologies permettant d’estimer la hauteur du couvert à partir d’observations par drone ou par robot roulant au sol. La faisabilité de deux principales technologies et plateformes ont été comparées et prouvées: le LiDAR porté par un véhicule au sol et des images RVB acquises par drone. Les deux chapitres suivants adressent le problème de l’estimation de la densité d’épis et de tiges de blé par images à haute résolution spatiale. Les résultats montrent le potentiel et les limites de l’apprentissage profond pour ces applications. L’accent est aussi mis sur l’étude des différentes configurations d’acquisitions possibles et le débit de la méthode. Le dernier chapitre revient sur les principaux résultats élaborés au cours de cette thèse et ouvre différentes perspectives pour le phénotypage haut-débit en remplacement ou complément des mesures manuelles classiquement réalisées par les sélectionneurs et propose des pistes pour améliorer les méthodes développées<br>Genetic progress is one of the major leverage used to increase food production and satisfy the needs for the increasing human population under global change issues. Selecting or creating the optimal cultivar for a given location is quite challenging considering the very large spatial and temporal variability of the environmental conditions. Field phenotyping, i.e. the quantitative monitoring of crop state variables and canopy functioning, was recognized as the bottleneck to accelerate genetic progress and increase crop yield. This multidisciplinary study develops statistical and image processing methods to estimate the several structural traits of wheat to be applied to crop breeding. Further, this thesis was undertaken in the context of rapid hardware and software technological advancements illustrated by the increasing accessibility to UAV (Unmanned Aerial Vehicle) and UGV (Unmanned Ground Vehicle) platforms, the decreasing cost of processing units (GPUs, cloud computing) and the boom in the development of deep learning algorithms. This manuscript is divided into five chapters: The first chapter introduces the motivation behind the study as well as the current needs for high throughput phenotyping. A state of the art on phenotyping is also achieved by drawing attention to image processing methods and convolutional neural networks. The second chapter presents the development of methodologies for estimating the crop height. The feasibility of two main technologies and platforms were compared and proven: LiDAR mounted on a UGV and RGB (Red Green Blue) images acquired by a UAV. The next two chapters address the problem of estimating the density of wheat ears and stems from spatial high-resolution images. The results show the potential and limitations of deep learning for this application. Emphasis is also put on the study of the different possible acquisition configurations and the throughput of the method. The last chapter summarizes the pipelines developed and draws different perspectives of high throughput phenotyping to replace or supplement in-situ measurements as well as the improvement facilitated by the methods developed