Road Crack Detection Based on Improved UNet Network

With the continuous improvement of urban infrastructure, timely detection and maintenance of road cracks have become crucial tasks to ensure road safety and sustainable development. In order to enhance the detection accuracy of road crack detection, this paper proposes an improvement to the UNet network specifically tailored for road crack detection. Firstly, we introduce the attention mechanism (CBAM), which helps the model better handle variations in road cracks at different scales, thereby enhancing the network's representational capabilities. Secondly, we adopt the DCNv2 network architecture, incorporating learnable offsets and deformable convolutions to adapt convolutional operations more effectively to the non-rigid shapes of the targets, improving the model's perception of fine structures in the images. The proposed algorithm is tested on a publicly available dataset of concrete crack images. Experimental results demonstrate that compared to the original UNet network model, the proposed approach achieves a 3.7
^%
increase in recall and a 6.5% improvement in accuracy.