Thesis - Open Access
Master of Science (MS)
Electrical Engineering and Computer Science
This paper proposes a pig pose estimation operating with Region Proposal Network (RPN) of Mask Region based Convolutional Neural Network (Mask R-CNN) and Visual Geometry Group (VGG) Neural Network (NN). Object pose estimations generates from the associations of different key points. Key points could be explained as specific location of an object such as different joints of a human body or joints of different object. Hourglass network is one of a NN delivering key points of an object. Associating the different key points with the hourglass network results could be represented as instance-level detection . However, the instance-level detection shows a lack of accuracy on the results. This algorithm provides limitations on the accuracy because the pairwise association is not produced on individual pigs which means extra calculation must be handled to connect the different body parts. During the process of associating the body parts, parts from other pigs might be involved. Mask R-CNN presents a feature of Region Proposal Network (RPN) which categorizes distinct objects of an input image depending on the model trained . In this paper we introduce a method providing the pig pose estimation constructed from the Mask R-CNN’s masking results. 230 images were operated as a dataset. An average of 14 pigs appeared in each of the 230 images. The VGG Neural Network was utilized for classifying the pig standing up or laying down position with the masked RGB image extracted from the Mask R-CNN.
Number of Pages
South Dakota State University
In Copyright - Educational Use Permitted
Lee, Sang Kwan, "Pig Pose Estimation Based on Extracted Data of Mask R-CNN with VGG Neural Network for Classifications" (2020). Electronic Theses and Dissertations. 4098.