This paper proposes an improved YOLOv8s model for road defect detection, addressing the limitations of the original YOLOv8s in terms of detection accuracy and model size. The improved model incorporates several key enhancements: a modified C2f-Faster-EMA structure that replaces the traditional bottleneck, a SimSPPF module with ReLU activation for improved computational efficiency, and a Detect-Dyhead module with attention mechanisms to enhance detection accuracy and adaptability. Experimental results on the RDD2020 dataset show that the improved model achieves a 5.8% increase in average accuracy (mAP@0.5), a 22.33% reduction in model size, a 23.03% reduction in parameter count, and a 21.68% reduction in computational complexity compared to the original YOLOv8s. The model also demonstrates superior performance in real-time detection and is suitable for deployment on resource-constrained devices. Comparative experiments with other object detection models, including Faster R-CNN, YOLOv3, YOLOv5, and YOLOv6, confirm the improved model's effectiveness in detecting road defects with higher accuracy and lower computational requirements. The study highlights the importance of optimizing model architecture to achieve better performance in road maintenance and safety applications.This paper proposes an improved YOLOv8s model for road defect detection, addressing the limitations of the original YOLOv8s in terms of detection accuracy and model size. The improved model incorporates several key enhancements: a modified C2f-Faster-EMA structure that replaces the traditional bottleneck, a SimSPPF module with ReLU activation for improved computational efficiency, and a Detect-Dyhead module with attention mechanisms to enhance detection accuracy and adaptability. Experimental results on the RDD2020 dataset show that the improved model achieves a 5.8% increase in average accuracy (mAP@0.5), a 22.33% reduction in model size, a 23.03% reduction in parameter count, and a 21.68% reduction in computational complexity compared to the original YOLOv8s. The model also demonstrates superior performance in real-time detection and is suitable for deployment on resource-constrained devices. Comparative experiments with other object detection models, including Faster R-CNN, YOLOv3, YOLOv5, and YOLOv6, confirm the improved model's effectiveness in detecting road defects with higher accuracy and lower computational requirements. The study highlights the importance of optimizing model architecture to achieve better performance in road maintenance and safety applications.