CA-HRNet: A Novel Approach for Yak Image Segmentation in Complex Environments

Semantic segmentation of yaks in high-altitude environments faces challenges including morphological diversity, complex coat textures, and environmental variations. This research introduces the Progressive Coordinate Attention Embedding Strategy (PCAES), strategically integrating Coordinate Attention mechanisms within HRNet at critical transitional nodes. Three key innovations are presented: (1) Position-specific embedding theory for optimal CA integration through systematic feature learning analysis; (2) Dual spatial enhancement framework with pre-preservation before downsampling and reconstruction optimization before upsampling; (3) Computational-performance balance achieving superior accuracy while maintaining deployment efficiency. The method addresses yak-specific challenges through targeted spatial-channel modeling for morphological variations, directional encoding for coat pattern recognition, and environmental robustness enhancement. Evaluation on a plateau yak dataset demonstrates CA-HRNet achieves 81.65% MIoU with significant improvements in boundary precision. Comparative visualization reveals substantial improvements in texture delineation and background discrimination compared to baseline methods. The framework enables efficient deployment for remote high-altitude livestock monitoring applications.