SRD-YOLO: A YOLO Framework for Small-Target Pavement Distress Detection

Aiming at the problems of low accuracy of existing pavement detection models and poor identification of small target lesions, a model SRD-YOLO is proposed for small target lesion detection. firstly, for the problem that crack lesions are not obvious and difficult to be labeled on asphalt and other pavements, RFAConv is used to construct a sensory field feature extraction module, which enhances the network's expression of crack features; secondly, a four-layered scale feature fusion structure is used to add small target feature detection heads, which makes the network easier to capture small targets such as cracks and potholes; finally, experiments are conducted on the RDD2022 dataset, and the research results show that the proposed algorithm improves the detection accuracy of mAP50 by 2.7%, and the detection frame rate of FPS can be up to 117, which is in line with the demand of real-time monitoring, under the condition of the original network without increasing the network parameters.