A Road Surface Disease Identification Model Based on Visual Semantic Segmentation and Conditional Generative Adversarial Networks

In order to improve the accuracy of small sample road pavement distress recognition, the article proposes a road pavement distress recognition model based on visual semantic segmentation and conditional generative adversarial network. An adaptive histogram equalisation method with restricted contrast is used to enhance the high-precision pavement image. Multi-channel feature map extraction, response map computation and feature fusion are performed on the enhanced image to obtain the pavement disease salient map. A visual semantic segmentation framework fusing hybrid dilated convolution and channel attention network is established to segment the salient map. Based on the conditional generative adversarial network, a disease recognition model is established to complete the disease type recognition. The experimental results show that the F1 value of the proposed model for road pavement disease recognition results remains above 0.95, which improves the accuracy of pavement disease detection.