Detecting Foreign Objects on Transmission Lines via Improved YOLOv5 Optimized by Attention Mechanism

The presence of foreign objects can bring great harm to transmission lines and may lead to problems such as short-circuiting, breaking, and tripping. In serious cases, it can also lead to fires and other safety incidents. Currently, the YOLOv5-based foreign object detection algorithm has achieved certain results, but the detection accuracy is still low in complex scenes because the traditional image detection algorithms have difficulty in dealing with complex occlusions and lighting changes present in the scene, causing the loss of key information during feature extraction. To cope with the problem, we incorporate the attention mechanism and propose the YOLOv5s-SimAM method that is able to detect foreign objects on transmission lines. We introduce SimAM in the neck network, which differs from the existing channel/space attention module in that no additional parameters are required. The SimAM is designed with an energy function to compute 3D attention weights, and then the weights are combined with the original feature maps to strengthen the network's focus on the key information. The experiment results suggest that the proposed method performs effectively with regard to detection accuracy.