| Title |
Development of a 3D Point Cloud-Based Sow Weight Estimation Model Using Center Point-Guided Oval Segmentation |
| Authors |
김민준(Min-jun Kim) ; 조현종(Hyun-chong Cho) |
| DOI |
https://doi.org/10.5370/KIEE.2026.75.7.1544 |
| Keywords |
3D Point Cloud; Deep Learning; Dynamic Graph CNN; PointNet; Segmentation; Sow Weight Estimation |
| Abstract |
Accurate sow weight monitoring is essential for reproductive management and feeding strategies in commercial pig farms. Conventional weighing methods require direct handling, which is labor-intensive and stressful. To address this, we propose a 3D point cloud-based sow weight estimation framework that incorporates Center Point-Guided Oval Segmentation (CP-OS) for input refinement. Raw point clouds captured in farm environments often include irrelevant points such as floors and surrounding structures, distorting geometric features and degrading weight estimation accuracy. The proposed CP-OS isolates valid body regions and suppresses background noise, enabling more reliable geometric feature learning. We evaluated the framework using PointNet and Dynamic Graph CNN (DGCNN), and CP-OS improved performance for both backbones. Notably, PointNet combined with CP-OS achieved an average MAPE of 10.76% across three repeated trials, showing the best overall accuracy. These findings highlight the importance of geometric input refinement for robust 3D sow weight estimation in real farm environments. |