Pavement Defect Detection Based on Deep Learning

The existing methods for pavement defect detection currently suffer from issues such as low accuracy, poor real-time performance, and a high missed detection rate for small target defects. To address these challenges, we propose an enhanced pavement defect detection method based on the state-of-the-art YOLOv8s model with exceptional comprehensive performance. In this study, we leverage the optimized YOLOv8 convolutional neural network to swiftly and precisely identify longitudinal cracks, transverse cracks, alligator cracks, potholes, and other road surface defects. The approach aims to reduce false detections and missed detections while enhancing both accuracy and speed of detection. We improve the neck structure of the YOLOv8s model by refining the BiFPN used in the EfficientDet model, thereby boosting overall model accuracy. Furthermore, we introduce a GAM global attention mechanism after the SPPF module of the YOLOv8 model to further enhance object detection precision. Experimental results demonstrate a 1% increase in mAP value after these improvements from 87.2% to 88.2%. This method outperforms the original YOLOv8 model approach and effectively identifies road defects with efficiency.