The pothole is a common road defect that seriously affects traffic efficiency and personal safety. Road evaluation and maintenance and automatic driving take pothole detection as their main research part. In the above scenarios, accuracy and real-time pothole detection are the most important. However, the current pothole detection methods can not meet the accuracy and real-time requirements of pothole detection due to their multiple parameters and volume. To solve these problems, we first propose a lightweight one-stage object detection network, the AAL-Net. In the network, we design an LF (lightweight feature extraction) module and use the NAM (Normalization-based Attention Module) attention module to ensure the accuracy and real time of the pothole detection process. Secondly, we make our own pothole dataset for pothole detection. Finally, in order to simulate the real road scene, we design a data augmentation method to further improve the detection accuracy and robustness of the AAL-Net. The metrics F1 and GFLOPs show that our method is better than other deep learning models in the self-made dataset and the pothole600 dataset and can well meet the accuracy and real-time requirements of pothole detection.

Road surface defect detection can effectively reduce maintenance costs, which is a critical component in road structural health monitoring. However, existing methods often face challenges in the heavy computational parameters and high-accuracy detection, limiting their practical applicability in resource-constrained industrial settings. To alleviate this gap, we propose a Lightweight and High-accuracy Network (LHA-Net) for road surface defect detection, consisting of three sub-networks for feature extraction, feature fusion, and detection head. First, the proposed Direction-guided Global Feature-Aware Module (DGFM) and the proposed Heterokernel Local Feature-Aware Module (HLFM) are used in the feature extraction sub-net to extract global and local features while minimizing network parameters. Second, we propose an Asymptotically Weighted Aggregation Mechanism (AWAM) in the feature fusion sub-net, which efficiently merges detailed and semantic features through asymptotic multi-scale fusion and weighted fusion at multiple stages. Third, we propose a Lightweight Decoupling Head (LDH) in the detection head sub-net to extract target location and category information by emphasizing defect details in horizontal and vertical dimensions. Finally, to improve the generalizability, we propose the RDD-CC dataset extension of RDD2022 using road images collected by automobiles in China. Compared with the well-established lightweight YOLOv8n, LHA -Net achieves comparable or superior mAP@.5 scores with gains of +0.8%, +0.5%, and +0.3% on RDD-CC, RDD-SCM, and RDD-SCD datasets, respectively. Remarkably, LHA-Net does so with only 2.5M parameters (reduced by 16%) and 5.6 GFLOPs (reduced the computational load by 30%). The code and datasets are available at https://github.com/ZCZST01/LHA-Net .