Improved Manhole Cover Defect Detection Using an Enhanced Backbone Architecture

This paper presents an improved manhole cover defect detection method aimed at enhancing detection accuracy and efficiency. We replaced the backbone network in the traditional YOLO model with the MobileMamba architecture and incorporated the LocalWindowAttention attention mechanism into the backbone network, thereby enhancing the model’s ability to extract features at different scales. Experimental results show that the model has achieved significant improvements across multiple performance metrics, with precision ($\mathbf{P}$) of 0.975, recall (R) of $0.93, \mathrm{mAP} 50$ reaching 0.951, and $\mathrm{mAP} 50-95$ at 0.89. The model has 3,400, 715 parameters, 10.7 GFLOPs, and a size of 14.2 MB. These improvements not only increase detection accuracy but also achieve a good balance in computational efficiency, making the model suitable for real-time manhole cover defect detection in practical engineering applications.