Lightweight concrete crack detection for urban intelligent management and maintenance

The continuous growth of global infrastructure stock has elevated the importance of smart urban maintenance, with concrete crack detection emerging as a critical component for intelligent infrastructure management. To enhance detection efficiency in this domain, a lightweight deep-learning model named GSGAA-Yolo is proposed for concrete crack detection. Firstly, the backbone and neck networks were reconstructed using ghost convolution modules to streamline the network architecture. Then, a novel feature extraction module (GSAA-C3k2) was designed based on the slim-neck architecture, incorporating agent attention mechanisms to optimise the accuracy–efficiency balance. Finally, the SPPELAN module is introduced to strengthen multi-scale feature extraction capabilities through spatial pyramid processing. Experimental validation on public datasets demonstrated that the proposed GSGAA-Yolo achieved 88.2% mean average precision, outperforming the baseline YoloV11 model by 1.1%. Compared with the baseline, the optimised architecture reduced the parameter count by 24% and the computational load by 19% while maintaining comparable inference speed. Cross-dataset evaluation confirmed the model's robust generalisation and transfer learning capabilities, indicating high practical value for infrastructure maintenance applications.