Research on Insulator Defect Detection Algorithm Based on Dual-Sampling TWave-YOLOv9

In recent years, insulator defect detection has grown crucial for power system maintenance and safety, with deep learning algorithms widely applied. To enhance detection accuracy and robustness while preventing characteristic information loss, we propose a TWave-YOLOv9 algorithm based on double sampling. Upsampled transposed convolution boosts spatial resolution and sensitivity to small defects, while wavelet subsampling convolution maintains spatial resolution and insulator details. The model introduces minimal increases in parameters and computational cost. It achieves an mAP50 of ${9 8. 3 \%}$ and mAP50-95 of ${8 6. 6 \%}$, significantly outperforming the original YOLOv9.