Instance Segmentation Method Of Complex Rock Fragments During Tunnel Excavation

Timely recognition of rock fragments and their morphological sizes can help in adjusting excavation parameters during tunnel boring machine(TBM)tunneling.Traditional manual inspection highly relies on experiences and subjective judgements of human operators and conducting sieving tests is not real-time and energy consuming.Rock fragments in real-world images are often observed against a dark background,distributed with a high size diversity,complicatedly distributed and blocked by each other.To solve these problems,a novel deep-learning-based method is proposed in this study to achieve real-time on-site rock fragment recognition.The main contents and results are as follows:To address the low illumination quality of original rock fragment images under realworld tunnel excavation circumstances,an automatic image preprocessing module including homomorphic filtering and histogram equalization is designed to improve the illumination intensity and contrast of images so that vision detection and recognition are feasible.To detect complicated rock fragments in real-world situations,an instance segmentation model comprises the object detection subnetwork and semantic segmentation is proposed.The object detection subnetwork is designed based on a modified SSD model.The semantic segmentation subnetwork is designed based on a modified Unet model.To solve distributed with a high size diversity,complicatedly distributed and blocked by each other of rock fragment images,prior anchors,multilevel feature fusion and selfattention modules are utilized in the instance segmentation.The results show that 88% of rock fragments can be recognized,and the average Io U values reach 0.75.To solve the problem of complete recognition of the super huge rock fragments,a multi-scale fusion method is proposed.A post-processing method of obtaining the major and minor axis lengths for rock fragments is proposed based on the approximated minimal external rectangles.The results show that the predicted size distributions of rock fragments fit well with the ground-truth ones.In conclusion,this study can provide both visual recognition and statistical results for the size distribution of rock fragments during TBM tunnelling to assist the prediction of rock properties and adjustment of excavation parameters.