Automatic Detection for Road Voids from GPR Images using Deep Learning Method

To realize automatic recognition of road void distress, the deep learning-based object detection algorithm-mask region-based convolutional neural network (Mask R-CNN) was proposed to recognize void distress automatically from ground penetrating radar (GPR) images. The Mask R-CNN model was applied to the GPR image detection of road void distress. The three tasks of automatic identification of highway structure disease, obtaining the morphological characteristics of distress, and positioning of the cavity were completed. The test results showed that this model has good recognition accuracy, with AP
50
and F -score reaching 92.3% and 92.0%, respectively, far exceeding other mainstream detection models. The detection results can meet the needs of engineering applications.