Substation Equipment Defect Detection Based on Improved YOLOv9

Defect detection of substation equipment plays a vital role in the power system, which contributes to the safe operation of the power system. To accurately detect the substation equipment defects, we propose a detection method based on improved YOLOv9. To extract the key features of small targets in the image of substation equipment, an efficient multi-scale attention mechanism (EMA) based on cross-space learning is introduced at backbone network layer of YOLOv9. In addition, to improve the convergence and accuracy of YOLOv9, the Inner-SIoU bounding box regression loss function is introduced rather than the original CIoU bounding box regression loss function, which makes bounding box regression (BBR) faster and more accurate. And experiments are conducted on a dataset containing seven kinds of substation equipment defects. The experimental results show that the accuracy (P), recall (R), and mean average precision (mAP) of our method are 86.9%, 73.5%, and 77.4%, respectively, which are better than that of the baseline methods.