Cold Region Road Surface Defect Detection Using an Intelligent Machine Vision Method

Defect detection is crucial for timely maintenance of roads in cold regions. Manual methods of road defect detection are inefficient and costly. In response to these problems, this paper proposes an intelligent method for detecting road defects in cold regions using machine vision technology. This method is based on the YOLOv8 framework and is a specially designed algorithm suitable for the characteristics of road images in cold regions. First, we designed a new C2f structure inspired by self-calibrated convolution to enhance the ability to obtain information from feature maps. Second, we introduced the low-parameter triplet attention module in the neck structure to expand the receptive field and improve detection accuracy. Finally, we redesigned the detection head structure to enhance multiscale features while reducing computational costs. Experiments on cold region road data sets (RDD2022-Norway and EdmCrack600) demonstrate that our method significantly enhances road defect detection performance, outperforming other mainstream algorithms.