GSC-YOLOv5: An Algorithm based on Improved Attention Mechanism for Road Creak Detection

Pavement damage detection is of great significance for traffic safety and pavement maintenance. Traditional detection methods require a lot of time and cost, and there are many problems in image detection. Therefore, an improved YOLOv5 pavement damage detection algorithm, GSC-YOLOv5, is proposed. First of all, to solve the problem that small cracks are similar to pavement and difficult to detect, we introduced the self-designed Spatial Pyramid Pooling module into the backbone network of the original YOLOv5 model: GM-SPPF to enhance the feature extraction of the network and improve the detection ability of fine cracks. Secondly, to solve the problem that it is difficult to locate defects under special lighting conditions, ShuffleAttention, an improved lightweight attention module, is introduced to help the model accurately locate features and improve the accuracy of object detection. Finally, the upsampling mode of the original model is changed to the lightweight CARAFE upsampling, so as to reduce the number of parameters in the model. The experiment was carried out on RDD2022 data set. The results showed that the improved YOLOv5 model mAP
0.5
reached 54.6%, which was 3.3% higher than the Baseline, and the F1 score reached 57.5%. Our algorithm can detect and locate the defects on the outlet surface, saving a lot of manpower and material resources, which is of great significance for the evaluation of highway damage and its maintenance.