Research on automatic classification of muck in earth pressure balance shield based on deep learning network

For the Earth Pressure Balance Shield Machine, the muck extracted from the screw conveyor outlet can promptly reflect the current soil information at the excavation face of the tunnel. Accurately determining the category of muck in real-time helps us understand the soil conditions at the current excavation face of the tunnel, aiding in formulating rational excavation strategies for the shield machine. This paper, based on the characteristics of the muck in this project and the impact of each type of muck on the practical operation of the shield machine, conducts muck classification. We establish visual recognition criteria for muck and modify the ResNet50 model to construct our own deep learning muck classification model, named "Ours." Through comparative experiments with other models using muck videos, the "Ours" model demonstrates excellent muck classification capabilities, with an average classification accuracy ranging from 97.2% to 100%. Its ACC_top1 ranges from 97.3% to 100%. In the end, we adopt the "Ours" model as the automatic muck classification model for this project.