A Deep Learning‐Based Method for Particle Size Distribution Analysis From TBM Rock Muck Images

Rock muck is a direct result of rock mass fragmentation during tunnel boring machine (TBM) excavation, and its morphological characteristics reflect the rock mass quality to some extent, offering a new technical approach for rock mass quality inversion and excavation parameter optimization. To overcome the limitations of traditional image processing methods in complex construction environments, this paper proposes a TBM rock muck particle size analysis technique based on digital image processing and deep learning. The goal is to extract real‐time morphological parameters of rock muck via image analysis, providing a foundation for rock mass quality inversion and dynamic excavation parameter adjustment. An industrial camera and LED light source are used to build an image acquisition system for excavated material. A fusion model combining faster region‐based convolutional neural network (Faster R‐CNN) and U‐Net is used to segment the contours of excavated particles and generate the rock muck gradation curve. Application results show that the system can automatically identify and segment rock muck and generate its gradation curve. Compared to traditional segmentation techniques, the proposed system handles complex rock muck images more effectively, offering improved recognition accuracy and faster processing speed. This study further confirms the engineering value of image recognition technology in TBM geological perception, promoting the development of intelligent construction and automated tunneling.