Dissertations / Theses on the topic 'Feature behavior analysis'

The initiative of a sustainable material system needs to lower the environmental and economic impact of production processes and adopt new ways of synthesizing and re-using materials. Even though the current conventional manufacturing processes, such as powder metallurgy, casting, forging, and rolling, have already shown their excellent ability to manufacture a large variety of parts and efficiently yield high volume products. Nevertheless, there are still many obstacles in manufacturing metallic components, such as complicated process procedures, high-energy consumption, large material waste, and high machinery cost for reasons that the excess materials need to be removed and extra post-processing time needs to be taken to acquire desired shapes during the machining stage. Thus, finding innovative solutions for producing complex structures is becoming increasingly desirable in the industry. Innovative additive manufacturing (AM, also known as 3D printing) techniques have proved their capacity to manufacture metallic materials with designed complex shapes and tailored properties. The selective laser melting (SLM) is one of the most popular AM techniques, which has the ability to manufacture a wide range of metallic powders in a layer-wise manner and fabricate complex shapes without compromising dimensional accuracy. The toxicity, biocompatibility, corrosion resistance, and stress shielding effect are the key challenges for developing titanium biomaterials for orthopedic applications. Adding nontoxic alloying elements into titanium can solve the issues of toxicity and biocompatibility. One of the best solutions for minimizing the stress-shielding effect and prolonging implant lifetime is to tailor the modulus of implant materials closer to that of bones. Nb is a nontoxic alloying element and an excellent β phase stabilizer, which plays a significant role in reducing the elastic modulus and in improving the corrosion resistance of Ti-based alloys. Accordingly, obtaining a highperformance simple alloy by reducing the alloying elements and substituting toxic elements can facilitate the improvement of sustainability. Thus, the β-metastable Ti-Nb alloys with relatively low elastic modulus have been studied for orthopedic implants due to their high strength to weight ratio, excellent corrosion resistance, and high biocompatibility in the human body. In addition, the high reactivity of titanium with hydrogen and oxygen as well as the high melting points of titanium alloys make conventional manufacturing difficult and cost intensive. As such, the SLM provides an innovative solution to manufacture shape-complicated products in a building chamber under the flow of high purity argon gas to minimize oxidation. However, the availability, printability, and high cost of high-quality raw metallic alloy powder are the limits for the SLM process. The individual elemental powder is relatively cheap and easy to manufacture. Thus, the use of elemental powder mixture results in greater alloy choices as well as lower cost and wider commercial availability. The issues of resultant microstructural and chemical inhomogeneity of the produced parts using the powder mixture have been the major concerns and challenges in the field. Since the mechanical behaviors and chemical properties directly depend on the microstructural homogeneity and phase composition, an in-depth understanding of the effect of inhomogeneity is required. It is necessary to have further advances in manufacturing optimization to extend the benefit of low production costs. In particular, in-situ alloying prospects make SLM a potential route to use a powder mixture with near infinite chemical compositions to synthesize desired titanium alloys for broad applications. As such, synthesizing the proper titanium alloys using the SLM technique, minimizing defect formation, controlling phase composition, evaluating their properties, and investigating the performances of SLM-processed products could significantly advance the applications in various industries and academia. The aim is to apply the SLM technique to process titanium alloys for biomedical and industrial applications. The results help to improve the scientific understandings of the interrelation among alloy compositions, processes, microstructures, defects, properties, and deformation behaviors of 3D-printed parts. Chapter 1 introduces additive manufacturing (AM) has huge potential to realize new alloys with flexible design and easy manufacturing. Especially for the customized healthcare products and services, such as biomedical implants, prosthetics, and hip replacement. Titanium alloys have desirable properties for various applications. Combining additive manufacturing with affordable and biocompatible titanium alloys can further advance and benefit the healthcare industry. Accordingly, the objectives are to fabricate titanium alloys by SLM and to investigate the microstructure, mechanical performance, and corrosion properties. Chapter 2 overviews the type, utilization, and advantage of AM techniques, biomaterials, and titanium alloys. The SLM process can manufacture parts with high precision and superb asbuilt surface quality but relatively high residual stress due to the rapid cooling rate. The raw powder properties and processing parameters play important roles in the densification and mechanical property of built products. The physical factors in the melting process and simulation are shown to understand the melt pool characteristics and stability, which is the critical factor to a successful and desired part. The microstructure, mechanical properties, and corrosion performance of different titanium alloys are also reviewed in order to design the powder, understand the mechanism, and improve the properties. Chapter 3 shows insight into the manufacturing of a Ti-35Nb composite using SLM and post heat treatment. The results emphasize the capability of SLM to fabricate alloys from elemental powder mixtures, even suitable for those with a significant difference in melting point. It provides a significant advance in the understanding of the effect of microstructural inhomogeneity on the resultant mechanical and chemical properties. Heat treatment can further enhance the corrosion resistance of SLM-produced Ti-35Nb samples because the improved chemical homogeneity can facilitate the homogeneous formation of titanium oxides and niobium oxides. It presents a different method of synthesizing novel β-type composites at a relatively lower cost and in easy manufacture. Chapter 4 shows the microstructure, phase response, and mechanical properties of the SLM-fabricated Ti-35Nb using an elemental powder mixture with reduced Nb particle size and its heat-treated counterpart. The results provide significant advances in the understanding of the role of undissolved Nb particles, Nb-rich interfaces, and Ti-Nb-based β phases on the mechanical performance. The nanoindentation mappings provide direct evidence of the contribution of the different phase responses to overall mechanical properties. The Nb particle segregation zones have lower hardness and higher deformation compared to the Ti-Nb matrix. The as-SLMed Ti- 35Nb exhibits relatively high tensile yield strength (648 ± 13 MPa) due to the formation of dendritic β grains. However, the ductility is relatively low (3.9 ± 1.1%) as a result of the weak bonding of undissolved Nb particles within the matrix. The heat-treated counterpart shows a slightly lower yield strength (602 ± 14 MPa) but a nearly 43% increase in ductility (5.6 ± 1.9 %) due to the improved homogeneous Ti-Nb β phase. Chapter 5 shows the microstructure, phase composition, melt pool morphology, and mechanical properties of a prealloyed Ti-35Nb alloy manufactured using SLM and compares it to one produced using an elemental powder mixture. The SLM-processed Ti-35Nb from both feedstocks retained a high volume fraction of β phase due to adequate β stabilization by the Nb and the fast cooling of the SLM process; however, other phase compositions were quite different. The chemical heterogeneity and inhomogeneous microstructure of the SLM-produced sample from powder mixture are results of the fast cooling rate of the melt pool and the high difference of melting temperature and density between elemental powders. However, a uniform microstructure and chemical composition can be achieved in the SLMed prealloyed Ti-35Nb. The variances of powder morphology, density, and melting point between mixed powder and prealloyed powder induce different melt pool status, where the stability of the melt pool plays a critical role in the homogeneity and microstructure. The SLMed Ti-35Nb prealloyed powder samples present a slightly lower yield strength (485 ± 28 MPa) but higher plastic strain (23.5 ± 2.2 %). The excellent ductility has been attributed to the high homogeneity, strong interface bonding, and the existence of a large amount of β phase. Chapter 6 shows the understanding of the homogeneity effect on the coexistence of the acicular α″, β grains, and melt pool boundary for a homogeneous microstructure. It provides some new insight into the phase response and the effect of homogeneity on the SLMed Ti-35Nb alloy using prealloyed powder. The reduced elastic modulus of β phase (89.6 ± 2.1 GPa) is close to that of α″ phase (86.3 ± 2.0 GPa) from the indentation measurement, which is in favor of orthopedic implants application. It also reveals that the nanoindentation test can provide a fast mapping and considerable potential to evaluate the homogeneity, microstructural features, individual phase strength, and deformation behavior in a fine microstructure of SLM-fabricated metallic alloys. Chapter 7 shows the preliminary design in porous structures and compressive behavior of different prealloyed Ti-35Nb sandwich composite porous structures manufactured using SLM. The simulation results were in good agreement with the compression tests. The compression tests show that the sandwich composites with different layers have different deformation behavior and mechanical properties. The rhombic dodecahedron porous structure with added layers could achieve balanced compressive strength and ductility. The preliminary sandwich design with the verified finite element method (FEM) models can be employed in other metallic porous structures to improve the strength and ductility without affecting the porosity. Chapter 8 concludes the present findings in this thesis and suggests the future challenges and development using SLM to tailor titanium alloys for specific applications. As such, the SLM technique is a promising route to develop titanium alloys from powder mixture with wider alloy choices at a cheaper cost and in easier availability. Even though a uniform microstructure and chemical composition can be achieved in the SLM-produced Ti-35Nb using prealloyed powder, there are still challenges on how to achieve full melting of elemental powder particles and obtain a homogeneous β phase microstructure. With the investigation of β- type Ti-Nb alloys, this thesis aims to further understand the effect of the unmelted Nb particles in the synthesized Ti-Nb alloys and melt pool stability as well as improve the Nb melting, microstructure, and mechanical properties for industrial and biomedical applications. Understanding the effect of powder feedstock type and phase features of the SLM-produced Ti- 35Nb using prealloyed powder further provides insights into the homogeneity, microstructure, and resultant properties. The novel design in Ti-35Nb sandwich composite cellular structures can benefit biomedical and industrial applications. By taking advantage of the commercial availability and lower cost of elemental powder, finding solutions to achieve full melting and homogeneous microstructure for nontoxic and biocompatible β-type Ti-Nb alloys with promising mechanical and corrosion properties is significant in future research and development.

The mission of the research presented in this thesis is to give computers the power to sense and react to human activities. Without the ability to sense the surroundings and understand what humans are doing, computers will not be able to provide active, timely, appropriate, and considerate services to the humans. To accomplish this mission, the work stands on the shoulders of two giants: Machine learning and ubiquitous computing. Because of the ubiquity of sensor-enabled mobile and wearable devices, there has been an emerging opportunity to sense, learn, and infer human activities from the sensor data by leveraging state-of-the-art machine learning algorithms. While having shown promising results in human activity recognition, most existing approaches using supervised or semi-supervised learning have two fundamental problems. Firstly, most existing approaches require a large set of labeled sensor data for every target class, which requires a costly effort from human annotators. Secondly, an unseen new activity cannot be recognized if no training samples of that activity are available in the dataset. In light of these problems, a new approach in this area is proposed in our research. This thesis presents our novel approach to address the problem of human activity recognition when few or no training samples of the target activities are available. The main hypothesis is that the problem can be solved by the proposed NuActiv activity recognition framework, which consists of modeling the hierarchical and sequential structure of human activities, as well as bringing humans in the loop of model training. By injecting human knowledge about the hierarchical nature of human activities, a semantic attribute representation and a two-layer attribute-based learning approach are designed. To model the sequential structure, a probabilistic graphical model is further proposed to take into account the temporal dependency of activities and attributes. Finally, an active learning algorithm is developed to reinforce the recognition accuracy using minimal user feedback. The hypothesis and approaches presented in this thesis are validated by two case studies and real-world experiments on exercise activities and daily life activities. Experimental results show that the NuActiv framework can effectively recognize unseen new activities even without any training data, with up to 70-80% precision and recall rate. It also outperforms supervised learning with limited labeled data for the new classes. The results significantly advance the state of the art in human activity recognition, and represent a promising step towards bridging the gap between computers and humans.

Emotion adds an important element to the discussion of how information is conveyed and processed by humans; indeed, it plays an important role in the contextual understanding of messages. This research is centered on investigating relevant features for affect classification, along with modeling the multimodal and multitemporal nature of emotion. The use of formant-based features for affect classification is explored. Since linear predictive coding (LPC) based formant estimators often encounter problems with modeling speech elements, such as nasalized phonemes and give inconsistent results for bandwidth estimation, a robust formant-tracking algorithm was introduced to better model the formant and spectral properties of speech. The algorithm utilizes Gaussian mixtures to estimate spectral parameters and refines the estimates using maximum a posteriori (MAP) adaptation. When the method was used for features extraction applied to emotion classification, the results indicate that an improved formant-tracking method will also provide improved emotion classification accuracy. Spectral features contain rich information about expressivity and emotion. However, most of the recent work in affective computing has not progressed beyond analyzing the mel-frequency cepstral coefficients (MFCC’s) and their derivatives. A novel method for characterizing spectral peaks was introduced. The method uses a multi-resolution sinusoidal transform coding (MRSTC). Because of MRSTC’s high precision in representing spectral features, including preservation of high frequency content not present in the MFCC’s, additional resolving power was demonstrated. Facial expressions were analyzed using 53 motion capture (MoCap) markers. Statistical and regression measures of these markers were used for emotion classification along the voice features. Since different modalities use different sampling frequencies and analysis window lengths, a novel classifier fusion algorithm was introduced. This algorithm is intended to integrate classifiers trained at various analysis lengths, as well as those obtained from other modalities. Classification accuracy was statistically significantly improved using a multimodal-multitemporal approach with the introduced classifier fusion method. A practical application of the techniques for emotion classification was explored using social dyadic plays between a child and an adult. The Multimodal Dyadic Behavior (MMDB) dataset was used to automatically predict young children’s levels of engagement using linguistic and non-linguistic vocal cues along with visual cues, such as direction of a child’s gaze or a child’s gestures. Although this and similar research is limited by inconsistent subjective boundaries, and differing theoretical definitions of emotion, a significant step toward successful emotion classification has been demonstrated; key to the progress has been via novel voice and visual features and a newly developed multimodal-multitemporal approach.

Foreground object detection is a fundamental task in computer vision with many applications in areas such as object tracking, event identification, and behavior analysis. Most conventional foreground object detection methods work only in a stable illumination environments using fixed cameras. In real-world applications, however, it is often the case that the algorithm needs to operate under the following challenging conditions: drastic lighting changes, object shape complexity, moving cameras, low frame capture rates, and low resolution images. This thesis presents four novel approaches for foreground object detection on real-world datasets using cameras deployed on moving vehicles.The first problem addresses passenger detection and tracking tasks for public transport buses investigating the problem of changing illumination conditions and low frame capture rates. Our approach integrates a stable SIFT (Scale Invariant Feature Transform) background seat modelling method with a human shape model into a weighted Bayesian framework to detect passengers. To deal with the problem of tracking multiple targets, we employ the Reversible Jump Monte Carlo Markov Chain tracking algorithm. Using the SVM classifier, the appearance transformation models capture changes in the appearance of the foreground objects across two consecutives frames under low frame rate conditions. In the second problem, we present a system for pedestrian detection involving scenes captured by a mobile bus surveillance system. It integrates scene localization, foreground-background separation, and pedestrian detection modules into a unified detection framework. The scene localization module performs a two stage clustering of the video data.In the first stage, SIFT Homography is applied to cluster frames in terms of their structural similarity, and the second stage further clusters these aligned frames according to consistency in illumination. This produces clusters of images that are differential in viewpoint and lighting. A kernel density estimation (KDE) technique for colour and gradient is then used to construct background models for each image cluster, which is further used to detect candidate foreground pixels. Finally, using a hierarchical template matching approach, pedestrians can be detected.In addition to the second problem, we present three direct pedestrian detection methods that extend the HOG (Histogram of Oriented Gradient) techniques (Dalal and Triggs, 2005) and provide a comparative evaluation of these approaches. The three approaches include: a) a new histogram feature, that is formed by the weighted sum of both the gradient magnitude and the filter responses from a set of elongated Gaussian filters (Leung and Malik, 2001) corresponding to the quantised orientation, which we refer to as the Histogram of Oriented Gradient Banks (HOGB) approach; b) the codebook based HOG feature with branch-and-bound (efficient subwindow search) algorithm (Lampert et al., 2008) and; c) the codebook based HOGB approach.In the third problem, a unified framework that combines 3D and 2D background modelling is proposed to detect scene changes using a camera mounted on a moving vehicle. The 3D scene is first reconstructed from a set of videos taken at different times. The 3D background modelling identifies inconsistent scene structures as foreground objects. For the 2D approach, foreground objects are detected using the spatio-temporal MRF algorithm. Finally, the 3D and 2D results are combined using morphological operations.The significance of these research is that it provides basic frameworks for automatic large-scale mobile surveillance applications and facilitates many higher-level applications such as object tracking and behaviour analysis.

Android is the second most targeted operating system for malware authors and to counter the development of Android malware, more knowledge about their behavior is needed. There are mainly two approaches to analyze Android malware, namely static and dynamic analysis. Recently in 2017, a study and well labeled dataset, named AMD (Android Malware Dataset), consisting of over 24,000 malware samples was released. It is divided into 135 varieties based on similar malicious behavior, retrieved through static analysis of the file classes.dex in the APK of each malware, whereas the labeled features were determined by manual inspection of three samples in each variety. However, static analysis is known to be weak against obfuscation techniques, such as repackaging or dynamic loading, which can be exploited to avoid the analysis. In this study the second approach is utilized and all malware in the dataset are analyzed at run-time in order to monitor their dynamic behavior. However, analyzing malware at run-time has known weaknesses as well, as it can be avoided through, for instance, anti-emulator techniques. Therefore, the study aimed to explore the available sandbox environments for dynamic analysis, study the effectiveness of fingerprinting Android malware using one of the tools and investigate whether static features from AMD and the dynamic analysis correlate. For instance, by an attempt to classify the samples based on similar dynamic features and calculating the Pearson Correlation Coefficient (r) for all combinations of features from AMD and the dynamic analysis. The comparison of tools for dynamic analysis, showed a need of development, as most popular tools has been released for a long time and the common factor is a lack of continuous maintenance. As a result, the choice of sandbox environment for this study ended up as Droidbox, because of aspects like ease of use/install and easily adaptable for large scale analysis. Based on the dynamic features extracted with Droidbox, it could be shown that Android malware are more similar to the varieties which they belong to. The best metric for classifying samples to varieties, out of four investigated metrics, turned out to be Cosine Similarity, which received an accuracy of 83.6% for the entire dataset. The high accuracy indicated a correlation between the dynamic features and static features which the varieties are based on. Furthermore, the Pearson Correlation Coefficient confirmed that the manually extracted features, used to describe the varieties, and the dynamic features are correlated to some extent, which could be partially confirmed by a manual inspection in the end of the study.

This thesis examines mental health professionals' responses toward patients diagnosed with Borderline Personality Disorder (BPD) and presents a qualitative study of Dialectical Behaviour Therapists' (DBT) experiences in their work with young people with BPD features. A review of empirical literature regarding emotional, behavioural and attitudinal responses of professionals toward these patients identified a range of negative responses, distinguishable from responses toward patients with other mental health problems. The review highlights the consistency of responses in professionals working in a variety of roles with these patients in countries across the world, and points to the need for further research to understand the precipitants of these negative responses. Controversy surrounds the diagnosis of BPD during adolescence and hence the majority of research in this area focuses upon professionals working with adult patients. On the basis of evidence regarding the presence of BPD features during adolescence and the application of therapeutic approaches, such as DBT, to young people exhibiting these features, the empirical paper presents an Interpretative Phenomenological Analysis (IPA) of the lived experience of DBT therapists in this context. A super-ordinate theme of 'the impact of the therapy on the therapist' containing five sub-themes is presented.

High hydrodynamic pressure (HDP) treatment is a novel non-thermal technology that improves tenderness in foods by subjecting foods to underwater shock waves. In this study non-destructive and destructive testing methods, along with two mathematical models were explored to predict biomechanical behavior of beef loins subjected to HDP-treament. The first study involved utilizing ultrasound and imaging techniques to predict textural changes in beef loins subjected to HDP-treatment using Warner-Braztler shear force (WBS) scores and texture profile analysis (TPA) features for correlation. Ultrasound velocity correlated very poorly with the WBS scores and TPA features, whereas the imaging features correlated better with higher r-values. The effect of HDP-treatment variables on WBS and TPA features indicated that amount of charge had no significant effects when compared to location of sample and container size during treatment. Two mathematical models were used to simulate deformational behavior in beef loins. The first study used a rheological based modeling of protein gel as a preliminary study. Results from the first modeling study indicated no viscous interactions in the model and complete deformation failure at pressures exceeding 50 kPa, which was contrary to the real-life process conditions which use pressures in the order of MPa. The second modeling study used a finite element method approach to model elastic behavior. Shock wave was modeled as a non-linear and linear propagating wave. The non-linear model indicated no deformation response, whereas the linear model indicated realistic deformation response assuming transverse isotropy of the model beef loin. The last study correlated small- and large-strain measurements using stress relaxation and elastic coefficients of the stiffness matrix as small-strain measures and results of the study indicated very high correlation between elastic coefficients c11, c22, and c44 with TPA cohesiveness (r > 0.9), and springiness (r > 0.85). Overall results of this study indicated a need for further research in estimating mechanical properties of beef loins in order to understand the dynamics of HDP-treatment process better.
Ph. D.

Za posledné roky je zaznamenaný nárast prác zahrňujúcich komplexnú detekciu malvéru. Pre potreby zachytenia správania je často vhodné pouziť formát grafov. To je prípad antivírusového programu Avast, ktorého behaviorálny štít deteguje škodlivé správanie a ukladá ich vo forme grafov. Keďže sa jedná o proprietárne riešenie a Avast antivirus pracuje s vlastnou sadou charakterizovaného správania bolo nutné navrhnúť vlastnú metódu detekcie, ktorá bude postavená nad týmito grafmi správania. Táto práca analyzuje grafy správania škodlivého softvéru zachytené behavioralnym štítom antivírusového programu Avast pre proces hlbšej detekcie škodlivého softvéru. Detekcia škodlivého správania sa začína analýzou a abstrakciou vzorcov z grafu správania. Izolované vzory môžu efektívnejšie identifikovať dynamicky sa meniaci malware. Grafy správania sú uložené v databáze grafov Neo4j a každý deň sú zachytené tisíce z nich. Cieľom tejto práce bolo navrhnúť algoritmus na identifikáciu správania škodlivého softvéru s dôrazom na rýchlosť skenovania a jasnosť identifikovaných vzorcov správania. Identifikácia škodlivého správania spočíva v nájdení najdôležitejších vlastností natrénovaných klasifikátorov a následnej extrakcie podgrafu pozostávajúceho iba z týchto dôležitých vlastností uzlov a vzťahov medzi nimi. Následne je navrhnuté pravidlo pre hodnotenie extrahovaného podgrafu. Diplomová práca prebehla v spolupráci so spoločnosťou Avast Software s.r.o.

This thesis investigates profiling and differentiating customers through the use of statistical data mining techniques. The business application of our work centres on examining individuals’ seldomly studied yet critical consumption behaviour over an extensive time period within the context of the wireless telecommunication industry; consumption behaviour (as oppose to purchasing behaviour) is behaviour that has been performed so frequently that it become habitual and involves minimal intentions or decision making. Key variables investigated are the activity initialised timestamp and cell tower location as well as the activity type and usage quantity (e.g., voice call with duration in seconds); and the research focuses are on customers’ spatial and temporal usage behaviour. The main methodological emphasis is on the development of clustering models based on Gaussian mixture models (GMMs) which are fitted with the use of the recently developed variational Bayesian (VB) method. VB is an efficient deterministic alternative to the popular but computationally demandingMarkov chainMonte Carlo (MCMC) methods. The standard VBGMMalgorithm is extended by allowing component splitting such that it is robust to initial parameter choices and can automatically and efficiently determine the number of components. The new algorithm we propose allows more effective modelling of individuals’ highly heterogeneous and spiky spatial usage behaviour, or more generally human mobility patterns; the term spiky describes data patterns with large areas of low probability mixed with small areas of high probability. Customers are then characterised and segmented based on the fitted GMM which corresponds to how each of them uses the products/services spatially in their daily lives; this is essentially their likely lifestyle and occupational traits. Other significant research contributions include fitting GMMs using VB to circular data i.e., the temporal usage behaviour, and developing clustering algorithms suitable for high dimensional data based on the use of VB-GMM.

碩士
國立中央大學
資訊工程研究所
93
Traffic accidents occur frequently in modernized society so that the society has to pay a lot of cost. There are more than 60,000 vehicle accidents occurring each year in Taiwan. Among them, over ninety percent of accidents are caused due to the careless of drivers according to the statistical analysis.In this thesis, a surveillance-based system utilizing face detection, tracking, and hazardous status monitoring of drivers is designed. By monitoring the symptoms of non-concentration or fatigue of drivers, a warning signal can be issued in advance so as to preventing the occurring of accidents due to the lack of unawareness. The proposed system is composed of four main parts. The first part is face detection. The modified version of YCbCr color space is adopted to obtain raw skin color images. Edge smoothing operation is employed to remedy the erroneous judgments of extracted features. Moreover, logarithmic intensity difference method is devised to compensate partial illumination. These algorithms make the task of face detection more robust and have higher accuracy than known skin-color model. The second part is face tracking. Correlation operation is manipulated on current face and records the ones which can regulate the borders and increase the accuracy. Time complexity of the proposed method can be drastically decreased due to the performing of face tracking. The third part is feature inspection and marking. Our proposed method can not only promote the accuracy but also mark the exact positions of features. A novel triangular-based theorem is adopted to calculate the angles of features to determine whether the considered face is frontal or profile. Moreover, it can conquer the problems of different face sizes, varying lighting conditions, varying expressions, and noises. The forth part is the analysis of dangerous behaviors. According to statistical results conducted by the Ministry of Communications, different weighs are assigned for five hazardous events which may result in accidents. The proposed system can automatically analyze the hazardous behaviors of chatting, drowsing, phone using, consecutive head lowering, and facial occlusion by performing direction estimation, pupil detection, feature variation, etc. Experiments were conducted on a variety of testing video sequences. An approximately 91% success rate can be achieved; besides with both false rejection rate and false acceptance rate being very low (near 10%). Experimental results reveal the feasibility and validity of our proposed system in monitoring various hazardous behaviors resulting from drivers.

博士
國立臺灣科技大學
電機工程系
103
The explosive amount of malware continue their threats in network and operating systems. Signature-based method is widely used for detecting malware. Unfortunately, it is unable to determine variant malware on-the-fly. On the hand, behavior-based method can effectively characterize the behaviors of malware. However, it is time-consuming to train and predict for each specific family of malware. We propose a generic and efficient algorithm to classify malware. Our method combines the selection and the extraction of features, which significantly reduces the dimensionality of features for training and classification. Based on malware behaviors collected from a sandbox environment, our method proceeds in five steps: (a) extracting n-gram feature space data from behavior logs, (b) building a support vector machine (SVM) classifier for malware classification, (c) selecting a subset of features, (d) transforming high-dimensional feature vectors into low-dimensional feature vectors, and (e) selecting models. Furthermore, we propose a Multi-Grouping algorithm for each feature reduction method. During the feature selection and extraction process, we show a easy way to figure out the major behaviors for each malware type. Experiments were conducted on a real-world data set with 4,288 samples from 9 families. As a proof of concept, we have evaluated our method by online training simulation experiment. Our 2-stages dimensionality reduction approach could have reduced the time cost significantly. The combination of MG TF-IDF, PCA and SVM for online training can finish the re-training and classifying in seconds, is sufficient to meet the online learning requirement for collecting the malware behavior in every minute. The experiments were demonstrated the effectiveness and the efficiency of our approach.

Dissertação de Mestrado Integrado em Engenharia Electrotécnica e de Computadores apresentada à Faculdade de Ciências e Tecnologia da Universidade de Coimbra
Esta obra insere-se no projecto de investigação ”Project Hometech”. Este focase num sistema de detecção de anomalias que pode monitorizar e detectar padrões de comportamento de pacientes com autismo. O sistema possibilita a detecção de estereotipias gravadas separadamente em vídeos offline e a previsão de resposta de um paciente a um determinado estímulo. As estereotipias constam numa lista de comportamentos exibidos recentemente ou de actividades previamente executadas. Com os dados recolhidos e a informação processada, é possível que essa base de dados possa também ser utilizada para tratar outros pacientes com semelhantes incapacidades [1]. A doença de autismo pode manifestar-se de vários sintomas possíveis tais como o afastamento extremo, falta de interacção social, comportamento repetitivo e violento entre outros. O desenvolvimento de terapia comportamental é relativamente desafiante uma vez que cada paciente autista é um caso isolado devido à grande variedade de sintomas e casos de intensidade. Usando o conhecimento obtido e analisando os dados sobre as actividades diárias do paciente poderá revelar padrões que ligam essas actividades. Isso poderá proporcionar aos terapeutas algum conhecimento prévio de prováveis resultados comportamentais possíveis relacionados com as suas terapias. Uma estrutura de reconhecimento de gestos foi criada com base num descritor local de movimento (LMD) com todas as informações necessárias [77]. A principal contribuição é propor um esquema de aprendizagem-classificação baseado em acompanhamento fiável através de pontos de interesse detectados utilizando o algoritmo Kanade-Lucas-Tomasi (KLT) em conjunto com o algoritmo mínimo de valores próprios desenvolvido por Shi-Tomasi [72]. Uma base de dados é gerada a partir do centroide, desvio padrão e velocidade média dos pontos de interesse acumulados. Na etapa final comparamos a base de dados gerada com uma sequência de testes com vários movimentos conhecidos e desconhecidos usando múltiplas máquinas de suporte de vector (SVM) binárias
This work is part of a research project called “Project Hometech”. It focuses on an anomaly detection system, which can monitor and detect behavior patterns of autism patients. Using separately recorded behavioral patterns in offline video footage, the system could predict the response of a patient to a stimulus, given a list of recently displayed behaviors or completed activities. The knowledge thus gathered could also be used to treat other patients of similar disability [1]. The autism disease can manifest itself by a wide variety of possible symptoms such as extreme withdrawal, lack of social interaction, repetitive and violent behavior between others. The development of behavioral therapy is relatively challenging since every autistic patient is an isolated case because of the wide variety of symptoms and intensity case. Collecting that knowledge gathered regarding the patient and analyzing the data about a patient’s daily activities could yield patterns linking these activities. This could thereby provide therapists with some foreknowledge of likely possible behavioral outcomes related to their therapies. A gesture recognition framework was created based on a Local Motion Descriptor (LMD) with all the necessary information [77]. The main contribution is to propose a learning-classification scheme based on reliable tracking of detected features using Kanade-Lucas-Tomasi (KLT) algorithm combined with the Minimum Eigenvalue algorithm developed by Shi-described in [72]. A database is created using the centroids, standard deviation and mean velocity of the clustered moving points. In the final step, we compare and classify the generated database with a test sequence through known and unknown stereotyped movements using multiple binary Support Vector Machines (SVM).

博士
國立成功大學
電機工程學系碩博士班
93
In recent years, Biomedical Informatics has become a new trend of science and technology. Biomedical Informatics is bringing together researchers from bioinformatics, medical informatics and computer science. The principal of this subject is using mathematical computation, statistics and computer analysis for life sciences research. Therefore, their application is very extensively, including gene, medical treatment, medicine and so on. 　In this dissertation, we focus on the auditory cell’s triggering feature analysis and behavior modeling. For triggering feature analysis, we proposed a method for determining the triggering features with accurate FM slope in STRF. For the behavior modeling, we proposed a novel system to simulate responses of auditory neurons of rats to acoustic signals. 　It is well known that speech sounds contain two major time-varying components, FM and AM signals. It is also clear that many FM-sensitive neurons appear in significant proportions first at the auditory midbrain and subsequently at the auditory cortex. To understand the mechanisms of speech sound coding, detailed knowledge on the triggering features of FM sensitive neurons is of obvious importance. Sensitivity of central auditory neurons to frequency modulated (FM) sound is often characterized based on spectro-temporal receptive field (STRF), which is generated by spike-trigger averaging a random stimulus. Due to the inherent property of time variability in neural response, this method erroneously represents the response jitter as stimulus jitter in the STRF. To reveal the trigger features more clearly, we have implemented a method that minimizes this error. 　We also propose a computational model to simulate central auditory responses to complex sounds. It consists of a multi-scale classification process, and an arti-ficial neural network composed of two modules of finite impulse response (FIR) neural networks connected to a maximum network. Electrical activities of single auditory neurons were recorded at the rat midbrain in response to a repetitive pseudo-random frequency modulated (FM) sound. The multi-scale classification process divides the training dataset into either strong or weak response using a multiple-scale Gaussian filter that based on response probability. Two modules of FIR neural network are then independently trained to model the two types of responses. This caters for the possible differences in neuronal circuitry and transmission delay. Their outputs are connected to a maximum network to generate the final output. After training, we use a different set of FM responses collected from the same neuron to test the performance of the model. Two criteria are adopted for assessment. One measures the matching of the modeled output to the actual output on a point-to-point basis. Another measures the matching of bulk responses between the two. Results show that the proposed model predicts the responses of central auditory neurons satisfactorily.

碩士
國立中央大學
資訊工程研究所
94
Due to the fast development of computer and video technologies and the cost-down of capturing devices, surveillance systems are gradually widely applied in our daily life. However, the main function of current surveillance systems only focuses on the recording of video event. The developing of automatic and intelligent surveillance systems can detect, track, recognize and analyze moving objects including the behaviors of objects and the occurring of abnormal events, and then issue warring message automatically. In this thesis, a video surveillance system for abnormal pedestrian behavior analysis is presented. Firstly, background subtraction technique is employed to detect moving objects from video sequences. Then, three key features, including object position, object size, and object color, are extracted to track each detected object. After that, Gaussian Mixture Models (GMM) is introduced to model pedestrians’ behaviors. According to the parameters of the models, different behaviors like walking, running and falling can be successfully recognized and analyzed. Finally, two curve-matching algorithms are employed to complete the trajectory retrieval. Experimental results show that the proposed method offers great improvements in terms of accuracy, robustness, and stability in the analysis of object behaviors.

碩士
國立中正大學
資訊工程所
96
Most human behavior systems nowadays use single feature or single classifier algorithm to do human behavior analysis. But if we use only one feature to analyze human behavior, many analysis can’t be do well in many case. For example, if we only use shape silhouette information to do human behavior analysis, the change information between frames in continue time doesn’t be considered in analysis procedure. And we also can’t realize what is the best analysis algorithm in some specific environment. In this paper, we use AdaBoosting based algorithm, with multiple features and different classifier, to combine all of them, and to create more reliable analysis results. For using AdaBoosting, one the restriction is all classifier must be weak. In this paper, we unanalyzed some actions deliberately to create many weak classifiers from original classifier. Finally, we combine all of them using AdaBoosting. The accuracy rate of these classifier is much lower than general classifiers, but after combing them by AdaBoosting algorithm, we will find that the better analysis results can be get in the experiments.

碩士
國立中正大學
資訊工程所
96
Existing human action recognition approaches can be classified into two categories: 2D based and 3D based. 3D based approaches may have high accuracy, whereas they use a lot of parameters to build model. 2D based approaches may have lower accuracy, whereas they are simple with low computational complexity. Thus, 2D based approaches are employed in most video surveillance systems. In this study, a human behavior analysis system using multiple features and a multicategory support vector machine is proposed. In the proposed system, three kinds of features, namely, human star skeleton, angles of six sticks in the star skeleton, and object motion vectors are employed to train the human posture classifier and recognize human postures. Based on the recognized human postures, a backward search strategy is proposed to recognize human actions. Based on the experimental results obtained in this study, the proposed system has good performance on human behavior analysis. In terms of recall and precision rates, the performance of the proposed system is superior to the comparison system.

碩士
中原大學
資訊工程研究所
96
With the advance of technology, the surveillance system has been widely used in many environments. Computer vision and multimedia techniques make feasible the development of new and “smart” surveillance systems that are different from traditional systems with only the recording function. In recent research, the surveillance system has been designed to automatically detect the location of human in video and to analyze the human behavior. In this paper, the objective is to identify and track the location of human, to match various static posture states in human behavior with respect to an established database, and to recognize the actual human behavior. The system design includes pre-processing of objects, object normalization, feature extraction of human, and collection of a database of human behavior for further analysis. The system was developed using 20 standard behavior as training samples, and was then evaluated with 32 unknown behavior samples. In addition, 2322 and 610 static poses were also evaluated. Our preliminary results demonstrated a 90% of classification in recognizing the two human behavior, namely the “walking” or “running”. In conclusion, human behavior is composed of many static posture states. Our methods could be used to analyze the combination of these states, and ultimately applied in surveillance systems with the need to recognize various humanbehavior.

In recent years, online shopping has gained immense popularity due to its feedback mechanism. By composing online comments, previous buyers share opinions and expe-riences regarding the items that they have purchased. These user-generated reviews, in turn, provide valuable information to potential customers in regards to deciding which products to purchase. The reviews also help vendors understand customer needs and improve product quality. Yet despite these benefits, the unprecedentedly rapid growth of user-generated content has overwhelmed human ability in online review scrutiny. On-line reviews that possess varying content further impedes useful knowledge distillation. The large volume of online reviews that are uneven in quality puts growing pressure on automatic approaches for effective review utilization and informative content prioritiza-tion. Review helpfulness prediction leverages machine learning methods to identify and recommend helpful reviews to customers. In particular, review characteristics form the backbone of helpfulness information acquisition. Prior literature has observed and as-sociated a large body of determinants with review helpfulness. However, these deter-minants heavily rely on the domain knowledge of experts. The selection of and the interaction between the determinants also remain understudied, leaving ample room for exploration. The general lack of systematic experiment protocols among the existing methods further harms the task’s reproducibility, comparability, and generalizability. This thesis aims to automatically model helpfulness information from online user- generated reviews. The thesis proposes effective modeling techniques and novel so-lutions to tackle the aforementioned challenges, with more emphasis on sophisticated feature learning and interaction. The thesis has made the following contributions to standardize the research field and advance the accuracy in helpfulness prediction. 1. A comprehensive survey is conducted to identify frequently used content-based determinants for automatic helpfulness prediction. A computational framework is developed to empirically evaluate the identified features across domains. Three selection scenarios are considered for feature behavior analysis. The domain-specific and domain-independent feature selection guidelines are summarized to facilitate future research prototyping. The implementation details of the study are discussed to standardize the task of automatic helpfulness prediction. 2. A deep neural framework is designed to enrich the interaction between review texts and star ratings during automatic helpfulness prediction. A gated convolu-tional component is introduced to learns content representations. A gated em- bedding method is proposed for encoding sophisticated yet adaptive rating infor- mation. An element alignment mechanism is proposed to explicitly capture the text-rating interaction. Ablation studies and qualitative analysis are conducted to discover insights into the interactive behavior of star ratings. 3. An end-to-end neural architecture is proposed to contextualize automatic helpful- ness prediction using review neighbors. Four weighting schemes are designed to encode a review’s surrounding neighbors as its context information into content representation learning. Three types of reviews neighbors of varied length are considered during context construction. Finally, discussions on the experimental results and the trade-o between model complexity and performance are given, along with case studies, to understand the proposed architecture.

碩士
國立中央大學
生命科學系
107
Valosin-containing protein (VCP, also known as p97) belongs to the family of type Ⅱ ATPase associated with variety of cellular activities (AAA). It functions as a chaperone involved in diverse cellular processes, including endoplasmic reticulum (ER) – associated protein degradation (ERAD), ubiquitin-proteasome system (UPS)-mediated protein degradation, ER and Golgi membrane fusion process. Mutations in the VCP gene cause multisystem disorders, such as inclusion body myopathy associated with Paget’s disease of bone and frontotemporal dementia (IBMPFD), amyloid lateral sclerosis (ALS), and autism spectrum disorder (ASD). Our previous study showed that Vcp and its cofactor p47 regulate ER morphology and protein synthesis efficiency in rat hippocampal neuron, and consequently control dendritic spine density. Leucine supplementation that promotes protein synthesis through mTOR pathway can rescue the reduction of spine density in Vcp+/R95G mutant neurons, suggesting the importance of protein synthesis in neuronal morphology controlled by Vcp. This thesis aims to further investigate the role of protein synthesis in VCP-regulated brain function. Here, I used Vcp+/R95G knock-in mice and wild-type littermates to conduct behavioral assays. Two kinds of diets with different protein content were provided to the mice. The results showed that no matter treated the mice with high or normal protein content diet, Vcp+/R95G mice showed comparable body weight and muscle strength to the wild-type mice, and had no defects in locomotion activity, spatial working memory, and rotarod tests. The anxiety level measured in light-dark box model was not increased in the Vcp+/R95G mice group. Nevertheless, we found that 2-month-old Vcp+/R95G mice fed with normal protein content diet reduced social interaction. In addition, the social deficit can be rescued by treating with high protein content diet. In this study, we demonstrate that raising the protein synthesis efficiency through the diet can improve the reduction of social tendency caused by Vcp+/R95G. These results suggested that protein synthesis plays an important role in regulating neuron function and social behavior, providing a potential therapeutic strategy for ASD treatment.

Restricted and repetitive behavior (RRB) is a core feature of autism spectrum disorder (ASD). Research suggests that the severity of RRB may be influenced by both environmental variables (e.g., absence of sensory stimulation input) and neurophysiological activity within the body (e.g., atypical regulatory capacity of the autonomic nervous system). Substantial research efforts have been devoted to the assessment of factors that influence the occurrence of RRB in individuals with ASD, which have led to the development of assessment methodologies, such as functional analysis, to identify specific contexts in which RRB occurs, and measures of heart rate variability (HRV) to index the level of neurophysiological activity for individuals with ASD.

However, despite the increasing consensus that the assessment and treatment of RRB require a more comprehensive approach due to the complexity and heterogeneity of the neurodevelopmental disorder, there exists a paucity in research that addresses both the functional behavioral and neurophysiological dimensions of RRB. This study aimed to address this gap by (a) designing and evaluating the effects of an integrated function-based assessment on identification of the functional subtypes of RRB and (b) examining the relationship between RRB and HRV as an indicator of neurophysiological functioning. The study included six participants, ages four to seven, with ASD. A single-case alternating treatments design, with two conditions simulating low- and high-stimulation environments, was used for the assessment of functional subtypes within each participant. Dependent variables included the duration of RRB and HRV. RRB was measured using MOOSES, a multi-option observation system for experimental studies. HRV was measured using wearable technology that collects blood volume pulse. Visual analysis of time series data as well as nonparametric analyses of the dependent variables were conducted to determine the functional subtypes of RRB and the association between HRV and RRB across participants.

Study results suggest that (a) the integrated assessment is effective in identifying specific functional subtypes of RRB and (b) HRV is positively correlated with the rate of RRB. The findings of this study offer new insights on the understanding of how underlying environmental and neurophysiological mechanisms may influence the occurrence of RRB in ASD. Furthermore, the study provides an integrated assessment model that can be feasibly implemented in applied settings.

abstract: With the rise of Online Social Networks (OSN) in the last decade, social network analysis has become a crucial research topic. The OSN graphs have unique properties that distinguish them from other types of graphs. In this thesis, five month Tweet corpus collected from Bangladesh - between June 2016 and October 2016 is analyzed, in order to detect accounts that belong to groups. These groups consist of official and non-official twitter handles of political organizations and NGOs in Bangladesh. A set of network, temporal, spatial and behavioral features are proposed to discriminate between accounts belonging to individual twitter users, news, groups and organization leaders. Finally, the experimental results are presented and a subset of relevant features is identified that lead to a generalizable model. Detection of tiny number of groups from large network is achieved with 0.8 precision, 0.75 recall and 0.77 F1 score. The domain independent network and behavioral features and models developed here are suitable for solving twitter account classification problem in any context.
Dissertation/Thesis
Masters Thesis Computer Science 2017

Smartphones are important means of communication that are rapidly advancing in technology. Its rather short replacement cycles continue to drive a growing volume of electronic waste. The introduction of a Repairability Index, which provides information on the reparability of an electrical device, is an attempt to extend useful life. By influencing consumers' purchase decisions, pressure is built up on manufacturers to enable their equipment to be repaired more cost-effectively and to provide replacement parts. The objective of this study is to investigate the influence of the Repairability Index on consumers' Willingness to Pay, Perceived Quality, and Purchase Intention. The importance of repairability is examined in the context of customer-preferred smartphone features to control for interaction effects. In a qualitative preliminary study, these preferred features were identified. Then, the main study was conducted using an experimental approach in which participants were presented with a manipulated smartphone advertisement. The results show that the Repairability Index influenced consumer purchase behavior by exerting a positive effect on Willingness to Pay and Perceived Quality. However, no effect was found on the overall Purchase Intention. This result is independent of age and gender, but stronger among individuals with high environmental awareness. Attitudes towards repair reinforced the effect of repairability on Perceived Quality. No interaction effect could be found between repairability and preferred features, however, preferred smartphone features revealed a great importance on all three variables in the experiment and therefore managers should not neglect them.
Os smartphones são importantes dispositivos de comunicação cuja tecnologia está a avançar rapidamente. Os seus curtos ciclos de substituição conduzem a um volume cada vez maior de resíduos electrónicos. A introdução de um índice de reparabilidade, que fornece informações sobre a reparabilidade de um dispositivo eléctrico, é uma tentativa de prolongar a sua duração de vida. Ao influenciar as decisões de compra dos consumidores, aumenta a pressão sobre os fabricantes para repararem os seus dispositivos de forma mais rentável e fornecerem peças sobressalentes. O objectivo deste estudo é investigar a influência do índice de reparabilidade na vontade de pagar, na percepção da qualidade e na intenção de compra por parte dos consumidores. A importância da reparabilidade é examinada no contexto das características preferidas dos smartphones dos clientes para controlar os efeitos de interacção. Foi realizado um estudo preliminar para identificar estas características preferidas. Depois, o estudo principal foi conduzido utilizando uma abordagem experimental na qual os participantes foram apresentados com um anúncio manipulado de smartphone. O índice influencia o comportamento de compra dos consumidores, exercendo um efeito positivo na vontade de pagar e na percepção de qualidade. No entanto, não foi encontrado qualquer efeito na intenção global de compra. Este resultado é independente da idade e do sexo, mas mais forte para indivíduos com elevada consciência ambiental. Não foi possível encontrar qualquer efeito de interacção entre a reparabilidade e as características preferidas, mas as características preferidas dos smartphones demonstraram uma grande importância nas três variáveis de experiência.