Study on an Improved YOLOv8 Algorithm for UAV Image-Based Road Crack Detection

Unmanned Aerial Vehicle (UAV)-assisted road crack detection is a key technology in modern road maintenance, which can effectively reduce the labor costs and improve the recognition efficiency compared to traditional detection methods. This study addresses the issues existing in current road crack detection methods and proposes an improved YOLOv8 model-based automatic recognition algorithm, YOLOv8-D-CBAM, suitable for high-resolution images collected by UAVs. Firstly, to more flexibly obtain crack information from the UAV perspective and enhance the recognition rate, we introduce depthwise separable convolutions into the backbone network of the model to improve its adaptability to complex scenes. Secondly, a Convolutional Block Attention Module (CBAM) is added to the neck of the model to further enhance the feature extraction capability in UAV images. With these improvements, our algorithm significantly increases the recognition rate of road cracks and the model’s performance. Experimental results show that the YOLOv8-D-CBAM algorithm achieves precision, recall, F1-score, and [email protected] of 98.21%, 98.21%, 97.9%, and 99.5%, respectively, on images collected by UAVs, with recognition rates for various crack diseases exceeding 90%. Therefore, the YOLOv8-D-CBAM algorithm demonstrates excellent model performance in the research of UAV-assisted road crack recognition.