Surface Feature Recognition of Surrounding Rock of TBM Tunnel Based on Mask R-CNN Model

In the construction of a TBM tunnel, the identification of surrounding rock surface characteristics is very important for the stability evaluation of surrounding rock and the selection of a support scheme. The traditional method mainly relies on manual observation, which is limited by the experience level of construction personnel and has obvious shortcomings. In this paper, a 3D reconstruction-based image acquisition method of tunnel-surrounding rock is proposed, and the Mask R-CNN model is used to identify, locate and segment the joint fissures and leakage characteristics of the surrounding rock surface. To overcome the defect of the original model in joint fissure identification, this paper improved the model by modifying the loss function and successfully solved the problem of joint fissure identification. The average accuracy of joint fissure identification was increased from 0 to 0.437, and the average accuracy of water leakage identification was increased from 0.400 to 0.662. Experimental results show that the optimised network structure has a significant improvement in both accuracy and speed and can realise high precision surrounding rock feature recognition.