TC-YOLOv8m: A Road Crack Detection Algorithm Based on YOLOv8m

To address the challenges of insufficient detection accuracy, excessive model complexity, and lack of crack-type discrimination in current road crack detection algorithms, this paper proposes an improved YOLOv8mbased detection algorithm named TC-YOLOv8m. The proposed method incorporates Haar Wavelet Transform (HWT) into the backbone network to reduce the number of parameters while enhancing multi-scale feature extraction capabilities. A lightweight TSA-C2f module is designed at the end of the backbone to improve the modeling of longrange dependencies and capture global context more effectively. Furthermore, a Coordinate Attention (CA) mechanism is embedded into the C2f structure of the neck, forming the CA-C2f module to strengthen the extraction of fine crack features and suppress background interference. Experimental results on a self-constructed road crack dataset demonstrate that TC-YOLOv8m achieves F1 score of 97.6 %, mAP50-95 of 64.4 %, and mAP50 of 99.0 %, which represent improvements of 1.4 %, 10.8 %, and 0.5 % respectively compared to the baseline YOLOv8m model. In addition, the model reduces parameters by 4.2 % and computational cost by 4.1 %. The algorithm achieves a detection speed of 19.2 FPS on <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$640 \times 640$</tex> resolution images, enabling real-time and accurate detection of various road cracks. These results demonstrate that the proposed TC-YOLOv8m algorithm offers a promising solution for intelligent and efficient pavement crack detection in practical applications.