The structural plane of surrounding rock is one of the key geological factors affecting the efficiency and safety of TBM excavation. Developing a rapid, precise, and adaptable method for identifying the structural plane of surrounding rock in TBM tunnels is of great significance. This article proposes a method for identifying rock structure planes based on multi camera vision and image recognition technology. Specifically, based on the equipment using binocular cameras, color and depth images of large exposed surrounding rocks are captured at fixed positions. The depth images are corrected to overcome image distortion issues based on positional parameters such as the position and shooting angle of the binocular camera. Furthermore, the Fine Boundary Description (CED) method was adopted to achieve precise identification of the structural planes of the surrounding rock. This approach is founded on traditional convolutional neural networks and adds a dual path of forward propagation and backward refinement of image data. In the backward refinement path, it continuously strengthens the capture of local boundaries in the image, captures the differences between structural planes and conventional surrounding rock pixels in the image, and then characterizes the boundaries of surrounding rock structural planes. Based on the TBM construction tunnel of Qingdao Metro Line 6, 427 sets of color and depth images of the surrounding rock were collected on site. By comparing the model recognition with the actual morphology of cracks, this method was further validated.