Lightweight Bird Detection Method for Substation Bird Strike Prevention

In recent years, with the improvement of the ecological environment and the advancement of power grid construction, the overlap between bird activity areas and substations or transmission lines has been increasing, leading to growing safety risks. To address the issues of traditional bird deterrent methods, which are time-consuming, labor-intensive, and ineffective, a dataset was first constructed by collecting bird images for model training. Next, the PicoDet model was applied to bird detection, and performance evaluation metrics for object detection models were proposed. Finally, the optimized model was validated through inference and compared with the performance parameters of different algorithms. By comparing with commonly used YOLO series object detection models, it was verified that the proposed PicoDet model not only achieves higher detection accuracy but also has the advantages of low memory usage and fast detection speed, making it suitable for high-precision real-time bird detection scenarios in substations.