Multi-scale distribution of coal fractures based on CT digital core deep learning

Abstract

In order to realize high-precision and high-efficiency identification of multi-scale distribution characteristics of coal fractures, carry out the study of multi-scale distribution characteristics identification methods based on CT digital core deep learning. Industrial CT scanning system is used to collect a large number of coal original CT digital core information array, the CT digital core information array is converted into a two-dimensional gray-scale image and then it is divided into square images of different scales and the image brightness is enhanced to different levels as training samples, Finally, the construction and optimization of model parameters of AlexNet, ResNet-18, GoogLeNet and Inception-V3 models for the identification of CT-containing fractures are realized by Matlab platform. Study the recognition accuracy and verification accuracy of different model training under different number of training samples; Study the accuracy, calculation efficiency and training time of different models for images with different scales and brightness under the same training sample, obtain the optimal model for calculating the fractal dimension of two-dimensional CT images with fractures, then, the fractal distribution characteristics of each fracture image are calculated according to the statistical method of box-counting dimension, compared with the traditional binarization method and human eye recognition method, The applicability of the multi-scale distribution characteristics identification method of coal fractures based on CT digital core deep learning is verified. The result shows: ① ResNet-18 model is the optimal model for calculating the fractal dimension of two-dimensional CT images with cracks when the image sample is brightness 4 and the scale is 3.5 mm to 21 mm, the model has high accuracy and short training time in calculating the fractal dimension of two-dimensional CT fracture images. ② Compared with the traditional binarization method, the multi-scale recognition method of coal fracture based on CT digital core deep learning has the advantages of fast speed, high accuracy and is not easily affected by impurities in coal.

Keywords

• coal fractures
• ct digital core
• image recognition
• crack identification
• ct scan