REU Student Project Abstract

Numerous methods of studying animal behavior have been developed and built upon through the help of technology. Segmentation provides the basics for analyzing and detecting movement changes, as well as behavioral. To structure the overview of this work, it's important to outline questions that may arise when doing segmentation: 1) what does this solve, 2) how are the findings analyzed, 3) and what can be done about hidden data. What motivates this research is if automatic detection of animal behavior can be helpful in developing decision support tools for the management of animals. Pigs were incorporated into the study as the animal of choice to test the competency of the machine learning device. Monitoring individual movement patterns helps researchers understand social dynamics and how the pigs habitat influences their experience. The approach described here includes training learning-models to use object recognition to identify pigs from images. There areapproximately800 images total that were labeled and will be applied in the learning models. The images were each labeled and the following methods were used to outline each individual pig: bounding boxes, key points, as well as polygon regional shapes. With bounding boxes, you set the image inside, key points you are building an outline by selecting the main points of the image, and the polygon allows at least 30 points to be placed along the image for a more accurate and fitting outline. These images are used to identify animals in different housing environments. The segmented images will be uploaded into a computer vision model to create a tool for automated animal tracking within a specific environment. These images can then be labeled using behavioral observations and used to ultimately create an automated behavior detection method.