An Improved Image Preprocessing Method for Defect Detection in Distribution Lines

In this paper, we propose an improved image preprocessing method to address the challenges of low resolution, cluttered backgrounds, and susceptibility to lighting, occlusion, and weather conditions in distribution line defect detection. Our method includes three main stages: noise reduction, contrast enhancement, and background segmentation with blurring. First, we analyze the types of noise in images and apply the Non-Local Means (NLM) denoising algorithm to minimize noise impact while preserving image details. Second, we utilize adaptive histogram equalization to enhance the contrast between power grid facilities and backgrounds, mitigating the effects of uneven lighting. Finally, we employ a foreground extraction method combined with Gaussian blur to obscure irrelevant backgrounds, thereby reducing interference from complex backgrounds. Experimental results demonstrate that our preprocessing method significantly improves the recall rate, mAPŀ50, and mAPŀ50:95 metrics of the YOLOv12 model in defect detection tasks. This provides a more robust foundation for defect detection in distribution line images and offers a reference for the intelligent inspection of distribution networks.