关键词: 隧道;掌子面节理;深度学习;神经网络;实例分割;模型试验;

Abstract: The current methods for detecting joints on tunnel face rely primarily on manual sketches, which has defects of low detection efficiency and subjectivity.To address these concerns, this paper presents an intelligent recognition and segmentation algorithm based on Mask R-CNN(Mask Region-based Convolutional Neural Network) for detecting joint targets on tunnel face images and automatically segmenting them, thereby improving detection efficiency and objectivity of the results.Additionally, to tackle the challenge of low detection accuracy in existing image processing methods, particularly for complex tunnel joint surfaces in dark environments, the paper introduces a Path Aggregation Network(PANet) to enhance the fusion capability of feature information in Mask R-CNN,thereby improving the accuracy of the intelligent detection method.The algorithm was trained on a dataset of 800 tunnel face images, and the research findings demonstrate that it can quickly detect the position of joints on tunnel face images and assign mask off code to the joint pixel regions to achieve joint segmentation.The mean average precision(mAP) of the detection boxes and segmentation in the 80 test set images were 58.0% and 49.2%,respectively, which outperforms the original Mask R-CNN algorithm and other intelligent recognition and segmentation algorithms.

Key words: tunnel; tunnel face joints; deep learning; neural network; instance segmentation; model test;