Research on the visualization method of lithology intelligent recognition based on deep learning using mine tunnel images

Abstract

An image processing and deep learning method for identifying different types of rock images was proposed. Preprocessing, such as rock image acquisition, gray scaling, Gaussian blurring, and feature dimensionality reduction, was conducted to extract useful feature information and recognize and classify rock images using TensorFlow-based convolutional neural network (CNN) and PyQt5. A rock image dataset was established and separated into workouts, confirmation sets, and test sets. The framework was subsequently compiled and trained. The categorization approach was evaluated using image data from the validation and test datasets, and key metrics, such as accuracy, precision, and recall, were analyzed. Finally, the classification model conducted a probabilistic analysis of the measured data to determine the equivalent lithological type for each image. The experimental results indicated that the method combining deep learning, TensorFlow-based CNN, and PyQt5 to recognize and classify rock images has an accuracy rate of up to 98.8%, and can be successfully utilized for rock image recognition. The system can be extended to geological exploration, mine engineering, and other rock and mineral resource development to more efficiently and accurately recognize rock samples. Moreover, it can match them with the intelligent support design system to effectively improve the reliability and economy of the support scheme. The system can serve as a reference for supporting the design of other mining and underground space projects.

1. Introduction

Rock recognition technology is a popular research topic in geology and ore resource exploration, and is primarily aimed at developing a preset method for identifying and classifying rock types by analyzing and processing rock sample images [1]. Early rock recognition methods were mainly based on manual observation and empirical judgment, and their accuracy and efficiency were lower owing to the influence of professional skills and subjective factors [2]. Rock image recognition methods primarily involve image-processing techniques based on feature extraction, including Fourier transforms, wavelet transforms, and color features [3].

Recent advancements in satellite and remote sensing technologies have facilitated the development of multicultural imaging methods for classifying various geological characteristics and rock types [4]. However, the cost of acquiring high-definition rock images in this domain using remote sensing techniques is high. Therefore, the substantial expense of data acquisition through hyperspectral methods using drones and satellites frequently hinders the automation of rock lithology identification. In contrast, rapid advances in image collection and computerized image pattern recognition techniques have led to the development of automated systems for identifying rocks from field-collected imagery [5–6]. Although artificial observation and empirical judgment technologies are less efficient, the use of satellite remote sensing technology for rock image recognition is more expensive. Accurate mastery of automatic systems based on deep learning algorithms to evaluate and identify rock lithology is one of the key links to ensure the successful application of underground engineering support.

Advancements in machine learning (ML) algorithms have resulted in major advances in rock recognition models [7]. Traditional convolutional neural network (CNN) models have various problems, such as overfitting [8]. In contrast, the application of a deep residual network (Reset) and an attention mechanism model (Attention) provide new pathways for solving these problems [9]. Zhao et al. [10] proposed a Reset-based system for rock image recognition. They used Dropout and L2 regularization to mitigate the overfitting problem and experimentally showed that this method substantially improved accuracy. The attention mechanism is an attention model that learns the circumstances under which a particular region is more important [11]. Zhang et al. [12] proposed an approach for rock image identification that utilized multiple attention mechanisms for rock recognition. They used diverse sizes of convolutional kernels and attention mechanisms to obtain different visual information, and attained favorable experimental outcomes. Lou et al. [13] investigated CNN-based rock classification and chemical element recognition in solidified rock cores and whole-rock images. They proposed a rock recognition method that combined deep learning and traditional chemical analysis. They achieved a higher rock classification accuracy by analyzing features, such as the spectral properties of different elements and rock textures. Wang et al. [14] utilized a multilayer perceptron machine model to classify basalt.