Tweet Share

• Send by email

  Send by email

  

  Automat ed Detection of Muck Types for EPB ShieldviaDeep Learning

  In the process of EPB shield tunneling, muck is a window to prob the soil condition just ahead of the tunnel face and engineers can keep abreast of the change of geologic condition accordingly. This paper presents a method for automated detection of muck types based on the images extracted from the on-site surveillance camera using deep learning algorithms. A practical muck classification method is proposed considering both the cone penetration test (CPT) data and engineering practice. The image data collected for this research are mainly from Shanghai area, so the shield muck discussed in this paper includes the following five types: soft clay, normal clay, hard clay, silty sand, and mixed soil. Vision features based muck recognition criteria about the five types are formulated for image labeling. A convolutional neural network (CNN) based on YOLOv4 is developed and applied to detect muck on a real case of Metro Line 14 in Shanghai. The experimental results reveal that the proposed muck classification system is reasonable and the YOLOv4 based detection model performs well in the automated detection of EPB shield muck and meet the real-time requirements.

  Recipient :

  Name ^* :

  E-mail address ^* :

  ^* Required fields

  Send   or  Cancel

• Print