Optimization algorithm for 3D bird target tracking in transmission corridor based on YOLOv8–GhostNet

High-voltage transmission lines enable long-distance power delivery but interact with ecosystems, as avian incursions pose risks to infrastructure. This study analyzes avian behavior to identify threats and implement warnings, reducing failure risks. To improve data accuracy, multi-object preprocessing and adaptive filtering are applied to corridor monitoring data. Edge-equalized local features extract multi-scale avian activity traits, forming a robust feature-mapping model. Focusing on avian incursions, the research uses image systems to gather data and develops a recognition framework combining YOLOv8 and GhostNet. The enhanced YOLOv8 algorithm addresses blind spots, improves intruder discrimination, and integrates parallel feature mapping for efficient cross-hierarchical processing. This expands feature space capacity, enabling stable avian detection in transmission corridors. Tests show the model outperforms YOLOv8 + Deep SORT and Faster R-CNN + SORT, increasing the MOTA metric by 6.92% and 8.25%, respectively, with better compatibility across embedded platforms.