Abstract: Ground Penetrating Radar (GPR) is widely used for detecting hidden road defects, but current detection methods for small-scale defects suffer from some problems including low accuracy and high missed detection rates. To address these issues, an improved YOLOv8 detection algorithm called SRM-YOLO is proposed. Firstly, the C2f module and StarBlock module are combined to design C2f_StarBlock to improve the computational efficiency of the model and optimize the feature extraction process. Secondly, the Receptive-Field Coordinate Attention Convolution (RFCAConv) is introduced to enhance the localization accuracy of the model to road defects areas by adaptively adjusting the receptive field and coordinate attention mechanism. Finally, Multi-scale Large Kernel Attention(MLKA) is used to replace Spatial Pyramid Pooling Fast (SPPF) module to improve the recognition ability of the model for small target defects and irregular defects. The experimental results show that the SRM-YOLO model achieves an improvement of 5.1% and 2.7% in mAP50: 95 and mAP50, respectively, compared to the baseline model on the constructed GPR image dataset for three common types of hidden road defects: cracks, water-rich areas, and voids. The experiment verifies the effectiveness of the proposed method.