A Pin-Missing Detection Method for Transmission Lines Based on Edge Information Enhancement and Fine-Grained Feature Extraction

Pins are crucial connecting components in transmission lines, and missing pins can lead to the detachment of fittings on the power tower. Due to the extremely small size of pins in drone inspection images of transmission lines, general object detectors are insufficient for effectively detecting missing pins. Therefore, this work constructs a cascade detection dataset and divides the detection task into two stages. The first stage is metal fittings detection, and the second stage focuses on pin-missing detection. Then, we propose a more effective auxiliary reversible branch (ARB) for YOLOv9 by incorporating dynamic snake convolution (DSC) into the generalized efficient layer aggregation network (GELAN) and employing efficient multiscale attention (EMA) for pin-missing detection. In addition, to further improve the accuracy of pin-missing detection, we combined the images’ edge information enhancement module (EIEM) with the improved YOLOv9. Experiments on our pin-missing dataset demonstrate that our method achieves significant improvements compared to the baseline model without affecting its inference speed. Moreover, compared to other state-of-the-art models, our method enables more accurate detection of missing pins.