A Railway Inspection System Based On A Semantic Segmentation Network And Image Processing
Ming‐An Chung, Chen-You Gao · 2024
Rail surface defects and missing fasteners are significant topics in railway detection. However, the detection of images may be subject to noise due to the diversity of environmental conditions, leading to a reduction in accuracy. Therefore, the paper proposes a method based on a semantic segmentation network and image processing. First, to address the problem of data preprocessing, this article introduces a bilateral filter into the U-Net algorithm to reduce noise and improve the efficiency of extracting edge features. Second, locations of rail surface defects and missing fasteners were detected from the U-Net. Finally, the article compares with other literature. The proposed method's effectiveness is demonstrated through experimental results, with accuracy rates of 99.9% for rail surface defects and 98.1% for missing fasteners, respectively.