Intelligent Recognition of Rare Birds that Accidentally Collided with Transmission Lines Based on Object Detection

To mitigate rare bird collisions with transmission lines and prevent resulting casualties and line tripping, this study proposes an intelligent bird recognition method based on object detection. An image dataset encompassing 11 rare bird species was compiled based on historical line collision data and field surveys conducted near transmission infrastructure. Image augmentation techniques, specifically fog simulation and noise addition, were applied to replicate real-world operating environments. The YOLOv10 architecture was improved by integrating a Large Separated Kernel Attention module into its feature extraction backbone. This modification reduces model parameters and accelerates bird feature extraction. Furthermore, aux heads were incorporated into both the feature extraction and fusion networks to strengthen the model’s capacity for learning distinctive bird characteristics, thereby boosting detection performance. Case study analysis demonstrates that the optimized model achieves a mean Average Precision (mAP) of 95.34%, an F1 score of 91.76%, and operates at 136.78 Frames Per Second (FPS), enabling efficient and accurate identification of rare birds posing collision risks to transmission lines.