Investigation on road underground defect classification and localization based on ground penetrating radar and Swin transformer

In response to the low detection efficiency and accuracy of traditional manual methods for detecting road underground defects, this paper proposes an intelligent detection method based on ground penetrating radar (GPR). This method integrates the detection, classification, and localization of road underground defects. The approach uses Swin Transformer as a feature extraction network and utilizes the YOLOX object detection algorithm as a road underground defect detection model. It enables the detection of defect regions in three types of defect images: voids, non-compact areas, and underground pipelines. In addition, the collected radar signals are processed by Fourier transformation to obtain time-domain spectra and frequency-domain spectra, which are further analyzed to generate signal classification data set to achieve the defect classification. Finally, based on the relative positional relationship between the detected defect images and the GPS information collected by the GPR, the real positions of the defects on the map are automatically determined using the APIs provided by Amap (AutoNavi map). Experimental results show that this method achieves a precision and recall rate of 94.2% and 99.1%, respectively, for the detection of road underground defects, with an average precision of 94% and an average classification accuracy of 90%. The method significantly improves the accuracy and speed of road underground defect detection while meeting engineering requirements, making it highly valuable for practical road underground defect detection tasks.