Academic literature on the topic 'Tongue Image Classification'

Automatic tongue image segmentation and tongue image classification are two crucial tongue characterization tasks in traditional Chinese medicine (TCM). Due to the complexity of tongue segmentation and fine-grained traits of tongue image classification, both tasks are challenging. Fortunately, from the perspective of computer vision, these two Tasks are highly interrelated, making them compatible with the idea of Multi-Task Joint learning (MTL).By sharing the underlying parameters and adding two different task loss functions, an MTL method for segmenting and classifying tongue images is proposed in this paper. Moreover, two state-of-the-art deep neural network variants (UNET and Discriminative Filter Learning (DFL)) are foused into the MTL to perform these two tasks. To the best of our knowledge, our method is the first attempt to manage both tasks simultaneously with MTL. We conducted extensive experiments with the proposed method. The experimental results show that our joint method out performs the existing tongue characterization methods. The process of tongue diagnosis by extracting meaningful features from tongue images and segmenting the relevant regions for analysis. The deep auto encoder neural network is employed to learn a compact representation of tongue images by encoding and decoding the input data.

In this work, we introduce a graphical interface for detection and classification of different tissue, focusing on tongue soft tissue, based on ADALINE neural networks to provide tools for a highly accurate diagnosis. The interface is capable to identify an affected area or even by exploration of an image of the same sample, to identify normal and pathological conditions. The Adaptive Linear Element (ADALINE) neural network successfully achieved a correct classification of 95% of total study cases, identifying either healthy or abnormal tissue, presented from a set of 70% of images for validations and 30% for training out of the total images.

The University of Georgia Department of Geography’s Center for Geospatial Research (CGR), with the support of the National Park Service (NPS) Vegetation Mapping Inventory (VMI) Program, described and mapped vegetation at Horseshoe Bend National Military Park (HOBE). This mapping effort was accomplished through collaboration with the NPS Southeast Coast Network (SECN), the North Carolina office of NatureServe (/Durham, N.C.), and Atkins North America, Inc. A final map of vegetation communities was created for Horseshoe Bend National Military Park (NMP) to the association level of the National Vegetation Classification System and in a 500-meter (1,640-foot [ft]) buffer zone around the park using the more general Anderson Level II classification. This map represents the vegetation found in the park during 2011, the year the images were acquired. We were provided with the vegetation communities occurring in the park, as determined by NatureServe from ground plots. We overlaid the location of the NatureServe plots on leaf-on color-infrared aerial photographs to determine the image signature of vegetation communities in terms of color, tone, texture, and topographic position. We also conducted our own field surveys to refine and verify photointerpretation. The park encompasses 829 hectares (2,049 acres [ac]) while the buffer alone covers 712 hectares (1,759 ac). Within the park boundary, there were 23 vegetation communities, with pine and hardwood forest communities dominating (88%). While forested, most of the forests (66% of total area) are mid-successional forests 30 to 75 years of age, reflecting past anthropogenic influences. The most common vegetation class is Early-to Mid-Successional Loblolly Pine Forest (24%). Areas impacted by exotic invasives or pine bark beetles were minimal (less than 1%). The buffer area (712 hectares [1,759 ac]) was 81% forested. There is a small component of rangeland (12%), which depending upon location and actual use, may influence water quality. Less than 3% of the buffer zone is high intensity anthropogenic land-uses. Because of the extensive past land use history, using the National Vegetation Classification System (NVCS) to the association level was challenging. Determining CEGL codes was particularly difficult because the National Vegetation Classification System was designed for relatively undisturbed vegetation communities. Using 16 NatureServe vegetation plots, color-infrared aerial photos, and data collected in the field, vegetation communities were delineated and assigned attributes. Using dominant vegetation classes and secondary vegetation classes, as well as modifiers to describe the diversity of species, detailed vegetation maps were created for Horseshoe Bend NMP. Polygons and attributes within the 500-meter (1,640-ft) buffer were created on a broader scale using a system based on the Anderson Level II classification scheme which includes anthropogenic and land use/land cover (LULC) classes. Within the park, the more detailed vegetation classes were used. The most common class found in Horseshoe Bend NMP is Early- to Mid-Successional Loblolly Pine Forest (6011) covering 24% of the park. This class combined with the second and third most common classes, Successional Sweetgum Floodplain Forest and Mid- to Late-Successional Loblolly Pine - Sweetgum Forest respectively, covers 57% of the park's 829 hectares (2,049 ac). The smallest class in the park that is larger than one half hectare is Highland Rim Pond (Woolgrass Bulrush—Threeway Sedge Type; 4719) which covers roughly one hectare. A rigorous accuracy assessment was conducted on the 23 map classes within the boundary of Horseshoe Bend NHP representing floristic types within the National Vegetation Classification System. Results showed 69% map accuracy and a kappa rating of 66% using 167 accuracy assessment points. The products generated from this project include vegetation maps, a photointerpretation key,..........