Automatic Cattle Body Measurement via YOLOv11 ‐ RSC and Depth Map Optimization

ABSTRACT

Manual measurement of cattle body size is time‐consuming, labor‐intensive, and relies on the experience of the staff. To address the limitations of traditional measurement methods, we present YOLOv11‐RSC, a lightweight framework built on YOLOv11n‐seg, and deployed it on edge devices. The model integrates the RepViT backbone, multi‐scale feature fusion, and lightweight attention mechanism to enhance segmentation performance under occlusion and complex backgrounds. Subsequently, the system completes regional division to locate key points and fuses incomplete depth maps with RGB images for depth map optimization. It constructs intact point clouds, projects two‐dimensional (2D) key points into three‐dimensional (3D) space, and finally realizes the automatic measurement of body height and other livestock morphological indicators. Experimental results indicate that the improved model achieved an increase of 1.1% in box mAP@50 and 0.8% in mask mAP@50 compared to the baseline, while reducing the number of parameters by 24.4% and the computational cost by 19.8%. The average relative error for these five cattle body measurements is less than 6%. Deployment has demonstrated the system's stability with a frame rate of 25.6 FPS. This study provides a practical solution for breeding management and the automatic extraction of body parameters.