Rock Thin-Section Image Classification based on Residual Neural Network

Abstract

When classifying rock slices, due to the small particle size of the rock slices, the classification is difficult. When manual methods are used for identification, the efficiency is low and subject to subjective factors. Therefore, this paper proposes a rock based on residual network The method of classifying granular images. This method uses the ResNet50 and ResNet101 models in the residual network to realize the automatic extraction of image features, and establishes a classifier to realize the classification based on the size of the rock slice image. This experiment uses 10,000 rock slice images obtained from the Ordos Basin, and uses 8,000 of them as the training set The residual network model is used for training, and then another 2,000 images are used to test the model. The experimental results show that two networks The accuracy of the classification results of the structure in the test set reached 90.24% and 91.63%. By using the residual network model to classify based on the rock slice image, an efficient and accurate classification effect can be obtained.