Rock Thermal Infrared Image Tension-shear Crack Detection Based On Deep Learning

The rock fracture can result in rock engineering disasters,and pose a major threat to mine production safety.The properties of cracks have a significant impact on the form of rock damage.It is critical to establish automatic classification and detection of tensionshear cracks for the prevention and control of mining disasters.In order to identify rock fracture patterns and monitor the failure process effectively,a network model capable of accurately detecting the property and location of tension shear cracks was trained using deep learning object detection methods based on the different thermal effects produced by different properties of cracks during the fracture process.The following are the important points.1)The uniaxial compression tests of granite and the direct shear tests of marble and limestone were carried out.From the process of rock fracture,526 infrared images of tension shear cracks were collected.These images are extended to 6422 images through data augmentation methods.Finally,they are made into a rock thermal infrared tensional shear crack dataset that can be used for deep learning object detection training.2)A rock thermal infrared image tension shear crack detection model based on improved Faster RCNN was proposed.This model is based on the Faster RCNN model with Res Net50 as the backbone network.To improve the base model,feature extraction optimization and detection frame regression optimization were utilized.Then a series of controlled experiments were set up.The results of experiments suggest that the introduction of attention-guided context feature pyramid network and GIo U loss can greatly increase model detection performance.The revised algorithm has a detection rate of 25.43 frames per second and a detection accuracy of 90.55 percent.Overall,the performance has improved.3)The temperature curve anomalies and infrared thermal image anomalies in the process of rock fracture were analyzed.The improved model was used to identify and detect infrared thermal images at typical time points during rock rupture.The results reveal that infrared abnormal characteristics of the rock can be detected before microcracks evolve into penetrating fractures.The findings suggest a method for forecasting the location and mode of rock fractures.Figure 40;Table 6;Reference 53...