AI-Powered DOLD-Net Enhances Livestock Detection Amidst Dense Occlusion
Researchers have developed DOLD-Net, a novel AI model designed for dense occluded livestock detection. This system, detailed in a recent study, significantly improves the accuracy of identifying livestock even when animals are closely grouped or partially hidden. DOLD-Net utilizes a Dual-Branch Occlusion-Aware Network (DBOAN) and a Context-Guided Focus Propagation (CGFP) framework to achieve its enhanced performance. The DBOAN component creates a robust feature representation by incorporating global topological information, addressing the limitations of local convolutions in processing densely distributed targets. The CGFP framework further refines feature representations and compensates for missing information in multi-scale features, improving efficiency. The model has been tested on various datasets, including SheepCounter and ChickenFlow, demonstrating strong generalization and robustness across different livestock species, occlusion levels, and environmental conditions. For instance, DOLD-Net-M achiev... 
