Recognition method of subway tunnel leakage diseases based on semantic segmentation

In order to accurately segment water leakage defects in subway tunnel lining images, a semantic segmentation improvement algorithm based on DeepLabV3+ was proposed. First, the lightweight network MobileNetV2 was used as the backbone network to effectively reduce the number of parameters; Second, the ASPP module was improved by connecting the atrous convolutions with different dilation rates to obtain rich contextual information; Then, the CBAM module was added to the coding layer structure to amplify the weight of the effective feature layer and improve the model's ability to perceive the water leakage region. The ResNet module was added to the decoding layer to fuse shallow and deep features to enrich the detail information and edge information. Finally, Focal Loss was used as the loss function to solve the problem of imbalance in the proportion of category pixels. The experimental results show that the F1, score and MIoU of the model on the test set reach 87.59% and 81.24%, and the improved model is able to recognize and segment the leakage disease effectively.