Segmentation of micro-cracks in fractured coal based on convolutional neural network

Abstract

The extension, penetration and expansion of coal and rock fracture structure caused by deep mining are the key factors that cause coal and rock damage.Therefore, the accurate characterization of the fracture structure is one of the core elements for understanding the coal and rock failure mechanism and developing coal and gas green co-mining.In this study, the Waifu2x convolutional neural network model was used to process the fractured coal and rock CT images to obtain CT images with higher resolution.And put forward the "seven-step processing method of fissure coal and rock labeling", which improves the efficiency of labeling, and uses image enhancement technology to expand the training set, which meets the model's requirements for the quantity and quality of the training set.The trained U-Net convolutional neural network is used for image segmentation to extract the fracture information in the coal and rock.Through comparison, it is found that the fracture connectivity, opening and distribution obtained by this method are closer to the real CT images, and the four quantitative indicators of the extracted fractures are better than other extraction methods.This study can provide a research foundation for the physical and mechanical behavior and mechanism analysis of fractured coal.

Keywords

• fractured coal
• image segmentation of micro-cracks
• convolutional neural network
• deep learning