Research on Rock Image Recognition Based on Deep Learning

Abstract

The classification and identification of rock samples is an important part of geological analysis. In order to reduce the cost of identification and prevent human subjective judgment from affecting the classification results, the use of deep learning to establish intelligent classification is a new way. Based on the deep convolutional neural network model of the AlexNet network structure, a deep learning migration model for rock image set analysis is established, and the automatic recognition and classification of rock lithology is realized by using the migration learning method. The traditional method of artificial image recognition of rock samples has big drawbacks. Based on the deep learning method, this paper establishes an image recognition model suitable for cuttings and core samples using industrial cameras at the logging site. Due to the small number of samples, the data is enhanced first. Then, feature extraction is performed on the rock image data based on a network structure such as a convolutional neural network. Finally, compare and analyze the experimental results of each model, and obtain the best model data for automatic identification of rock lithology. Thereby reducing the cost of rock image recognition and improving the efficiency of geological work.