Abstract
Road damage detection is of great significance to traffic safety and road maintenance. However, the existing target detection technology still has shortcomings in accuracy, real-time and adaptability. In order to meet this challenge, this study constructed SEA-YOLO v8 model for road damage detection. Firstly, the SBS module is constructed to optimize the computational complexity, achieve real-time target detection under limited hardware resources, successfully reduce the model parameters, and make the model more lightweight; Secondly, we integrate the EMA attention mechanism module into the neck component, enabling the model to utilize feature information from different layers, enabling the model to selectively focus on key areas and improve feature representation; Then, an adaptive attention feature pyramid structure is proposed to enhance the feature fusion capability of the network; Finally, lightweight shared convolutional detection head (LSCD-Head) is introduced to improve feature representation and reduce the number of parameters. The experimental results on the RDD2022 dataset show that the SEA-YOLO v8 model has achieved 63.2% mAP50. The performance is better than yolov8 model and mainstream target detection model. This shows that in complex urban traffic scenarios, the model has high detection accuracy and adaptability, can accurately locate and detect road damage, save manpower and material resources, provide guidance for road damage assessment and maintenance, and promote the sustainable development of urban roads.