Research on Pavement Distress Detection Method Based on Deep Learning

Pavement distress detection is a critical component of road maintenance, aiming to ensure road traffic safety and the quality of road maintenance. This paper presented a pavement distress detection algorithm based on deep learning that utilises ResNet18 to replace the original network's Darknet53. The use of residual connections effectively retains the underlying features of pavement distresses. Moreover, by adding the SeNet attention mechanism to the residual connections, it is possible to independently and adaptively adjust the weights of each channel of the residual connection's input feature map. This enables the network to capture critical features in the data better, thereby improving the model's performance. Experimental results on the Unmanned Aerial Vehicle Pavement Distress (UAPD) dataset show that the improved network's mean Average Precision (mAP) increased by 4.6% while maintaining the same detection speed, effectively enhancing the accuracy of pavement distress detection.