CS-Unet: a method for assessing the hazard of water ponding in railway drainage ditches

Abstract In view of the problems that the terrain and geological conditions at the portals of railway tunnels are complex, the routine inspection of the front slopes drainage ditches is rather difficult, and it is also hard to efficiently handle the water ponding problem, this paper proposes a method for evaluating the water ponding hazard of the front slopes drainage ditches in railway tunnels based on an improved Unet. Firstly, by adding the Convolutional Block Attention Module (CBAM) during each down - sampling process of the Unet model to enhance the segmentation accuracy of the model, and replacing the original backbone network ResNet (Residual Network) of the Unet model with the ShuffleNet v2 network to reduce the number of model parameters and improve the segmentation efficiency of the model, the CS - Unet model in this paper is constructed. Subsequently, this paper uses the self - built high - definition unmanned aerial vehicle (UAV) image dataset of railway drainage ditches to conduct sufficient ablation experiments and comparison experiments on the CS-Unet model. The experimental results show that, compared with the original model, with an improvement of 2.52%, 1.97%, and 2.18% in Accuracy , mPA , and mIoU metrics, respectively. Furthermore, the number of parameters is also reduced to 23.7% of that of the original model. This further confirms the effectiveness of the improvement scheme in this paper and the superiority of the model’s segmentation performance. Finally, based on the model segmentation results, a hazard assessment method for the proportion of water ponding area was constructed, and a quantitative assessment experiment was carried out on some railway drainage ditch water ponding image data to verify the feasibility of the method. Our proposed CS-Unet model provides a new idea for the accurate detection and scientific assessment of railway drainage ditch water ponding diseases and diseases in similar scenarios, which is of great significance for improving the overall efficiency of railway disease inspection and ensuring the safe and stable operation of the railway system.