Research on Pin Defect Detection Algorithm of Power Transmission Line

A pin defect detection method utilizing an enhanced YOLOv5 algorithm was introduced to enhance the efficiency of UAV inspection on overhead power transmission lines and bolster the detection rate of critical device defects within the lines. Aiming at the attributes of a sizable image dimensions and small detection target of UAV, a cascaded three-level network detection method is adopted. ConvNeXt Block was introduced to build CNeB Block and replace some C3 Block in Neck. The RepLKDeXt Block was constructed in combination with RepLKNet to replace C3 Block in the trunk and Neck, and a depth of 27×27 separable convolution was used to improve the effective receptive field and higher shape deviation. A dynamic convolution ODConv was added to replace the Neck part of the Conv, and all four dimensions of complementary attention were learned using the multidimensional attention mechanism and parallel strategy. The experimental results demonstrate that compared to the YOLOv5s network, the enhanced model achieves a [email protected]:0.95 of 70.5%, marking a 2.5 percentage point increase, with an [email protected] reaching 99.5%. Additionally, the model’s computational load is decreased by 1, parameters reduced by 4.77%, effectively enhancing the detection accuracy of power transmission line pins while also improving detection speed.