57 / 2025-04-14 15:00:41

Light-Robust Object Detection for Pig Suckling Behavior Monitoring in Variable Farm Illumination Conditions

Piglet suckling behavior,Deep learning,Object detection,Precision livestock farming

3. 智慧牧场与动物福利 > （4）智慧牧场应用的案例实践

Monitoring pig suckling behavior is essential for evaluating piglet welfare and productivity in modern swine production. Current computer vision methods face significant challenges from inconsistent lighting conditions in commercial farm environments. This paper presents a Light-robust Contrast Normalization (LCN) detection network that addresses these illumination variations. Our approach incorporates a novel architecture that enhances feature extraction under diverse lighting conditions through integrated contrast normalization and adaptive attention mechanisms. Evaluation across datasets of varying image quality shows that LCN achieves 93.3% precision, 93.8% recall, and 98.2% mAP50 while outperforming the YOLOv12 baseline with improvements of 1.7%, 13.9%, and 5.1% respectively. The LCN network demonstrates 22% lower performance degradation under poor lighting conditions compared to YOLOv12. Ablation studies confirm each architectural component's contribution to overall robustness. The enhanced performance of LCN enables reliable 24-hour monitoring of piglet suckling behavior, facilitating comprehensive welfare assessment and providing deeper insights into piglet nutrition patterns.