RIEC-YOLO: an improved road defect detection model based on YOLOv8

Detection of road defects plays a vital role in ensuring road safety. Existing road defect detection methods often struggle to simultaneously meet the requirements of accuracy and speed due to the diverse scales and complex backgrounds of road defects. This paper proposes an enhanced road defect detection model, RepViT-iEMA-CN2C2f-YOLO (RIEC-YOLO) network, based on YOLOv8. Firstly, to enhance the model's ability to learn contextual features in crack areas, a lightweight backbone feature extraction network, RepViT-M1.5, is used to replace the base network of YOLOv8. Secondly, to suppress irrelevant background information and reduce the probability of false alarms, a ConvNeXtV2-C2f (CN2C2f) module is designed to replace some C2f modules in the neck network. Meanwhile, to more effectively differentiate crack types, a novel inverted residual EMA (iEMA) attention mechanism module is proposed, which can extract features efficiently and fuse multiple scales. Finally, this paper validates the effectiveness of the proposed improvement methods through comparative experiments and ablation studies, and compares the RIEC-YOLO model with other state-of-the-art models. Compared to the YOLOv8x, the proposed model achieves a 1.4% improvement in mAP50 with only 16.9% of the computational cost. The performance significantly exceeds models such as YOLOv8x, demonstrating more competitiveness in efficient detection of road defects.