Instance Segmentation and Automated Pig Posture Recognition for Smart Health Management

• DOI: 10.5187/jast.2024.e112
• Publication Date: 2024-11-20T05:30:53Z
• Authors: Md Nasim Reza; Md Sazzadul Kabir; Md Asrakul Haque; Hongbin Jin; Hyunjin Kyoung; Young Kyoung Choi; Gookhwan Kim; Sun-Ok Chung

Changes in posture and movement during the growing period can often indicate abnormal development or health pigs, making it possible to monitor detect early morphological symptoms risks, potentially helping limit spread of infections. Large-scale pig farming requires extensive visual monitoring by workers, which is time-consuming laborious. However, a potential solution computer vision-based movement. The objective this study was recognize using masked-based instance segmentation for automated closed farm environment. Two automatic video acquisition systems were installed from top side views. RGB images extracted files used annotation work. Manual 600 prepare training dataset, including four postures: standing, sitting, lying, eating food bin. An framework employed posture. A region proposal network Mask R–CNN-generated candidate boxes features these RoIPool, followed classification bounding-box regression. model effectively identified standard postures, achieving mean average precision 0.937 piglets 0.935 adults. proposed showed strong real-time welfare issue detection aiding optimization management practices. Additionally, explored body weight estimation 2D image pixel areas, high correlation with actual weight, although limitations capturing 3D volume could affect precision. Future work should integrate imaging depth sensors expand use across diverse conditions enhance real-world applicability.