@article{JI2026, 
author = {Yongcheng JI and Yi LI and Hanping CHEN and Yang LIANG},
title = {A Lightweight Crack Detection Algorithm Based on Improved YOLOv11},
year = {2026},
journal = {Journal of South China University of Technology (Natural Science Edition)},
volume = {54},
number = {5},
pages = {47-58},
keywords = {road engineering, crack detection, lightweight object detection, attention mechanism, deep learning},
url = {https://www.sciopen.com/article/10.12141/j.issn.1000-565X.250158},
doi = {10.12141/j.issn.1000-565X.250158},
abstract = {Road pavements, as critical transportation infrastructure, are prone to cracks and other defects under long-term load bearing and environmental erosion. Traditional manual inspection methods suffer from low efficiency, high subjectivity, and significant safety risks, making it difficult to meet the maintenance demands of largescale pavement networks. This paper proposes a lightweight, high-precision pavement crack detection algorithm named LMC-YOLO (Lightweight Mobile-NetV4 with CAA for YOLO) that addresses feature loss and insufficient accuracy in detecting thin and elongated cracks. The algorithm systematically optimizes the backbone structure, attention mechanisms, and lightweighting strategies of the detection network. By introducing the lightweight Mobile-NetV4 (MNV4) structure into the backbone network and utilizing its efficient Universal Inverted Bottleneck (UIB) modules, the algorithm achieves a balance between powerful feature extraction capability and low computational cost. An improved Context Anchor Attention (CAA) mechanism is integrated into the neck of the detection network, along with a crack shape-aware module based on strip convolution, effectively enhancing the detection capability for elongated cracks. Through refined network design, the number of model parameters is reduced by 23.3%, computational complexity is decreased to 4.6 GFLOPs. Experimental results demonstrate that LMC-YOLO achieves 91.1%precision, 84.6% mAP@0.5, and 70.3% mAP@0.5∶0.95 on crack detection tasks, with an F1-score of 80.40%and an inference speed of 345 frames per second. Cross-dataset validation on DIOR and DOTA-v2 further confirms the model’s cross-domain transfer capability. This method successfully achieves an effective combination of high accuracy and efficient lightweight design, providing a practical solution for real-time pavement crack detection on mobile and embedded devices.}
}