A Lightweight Foreign Object Detection Method For Transmission Lines Based On The Improved YOLOv5n

Aiming at the technical bottleneck that detection accuracy, detection speed and lightweighting are difficult to optimize in the context of foreign object detection on transmission lines, this study proposes an improved You Only Look Once Version 5 Nano(YOLOv5n) detection method for edge computing devices. First, the Complete IoU (CIoU) loss function is replaced with Efficient IoU (EIoU), which effectively improves the regression accuracy between the prediction box and the real box. Secondly, the Bidirectional Feature Pyramid Network (BiFPN) architecture with bidirectional cross-scale connection is adopted to enhance the multi-scale feature fusion capability, and the Bidirectional Feature Pyramid Network with Concatenation (BiFPN_Concat) module is used instead of the conventional concatenation operation to achieve full fusion and utilization of multi-level features. Finally, the number of channels in the convolutional layer is scaled proportionally to compress the model volume. The results showed that while maintaining the lightweight advantage, the improved model improved the mean average precision (mAP) by 2.9%, the precision (P) and recall (R) by 7.5% and 2.5% respectively, and the detection speed by 15 frames per second (FPS) compared with the baseline model. The method performs well on transmission line foreign object detection devices equipped with edge computing platforms and has practical application value.