Comparison and Improvement Analysis of Coal Damage Feature Image Recognition Methods Under Loading Conditions

Abstract

With the transfer of coal mining to the deep part of China, the geostress environment endured by the deep coal body is more complicated, which seriously threatens the safe and efficient production of coal mines. This paper aims to investigate the accuracy of common image recognition methods in identifying coal cracks. It seeks to lay a foundation for understanding the macroscopic destruction mechanisms of coal bodies and the prevention and control of coal mine gas hazards by analyzing the evolution of macroscopic fissures in coal samples during loading experiments. High-speed video cameras recorded the coal body destruction process. Comparative analysis was performed on coal samples under four loading conditions using eleven image recognition methods in Matlab. The completeness and accuracy of each image were assessed based on the pixel counts of binarized images. The findings reveal that all methods, except for the region growth and mathematical morphology processing methods which lost significant crack information, effectively recognized crack information. The 2D convolution method exhibited excellent performance in image recognition, achieving the highest completeness and accuracy at all stages. After further optimization and enhancement, significant improvements in crack recognition were observed, with the completeness and accuracy of the Stage 1 coal samples increasing by 9.19% and 8.13%, respectively. Stage 4 coal samples had the highest checking completeness and accuracy rates, reaching 64.14% and 69.73%, respectively. The results of this paper provide a theoretical reference for the development of image recognition technology and safe coal mining.