Automatic Classification Method Of Rock Slices

Because of the complex composition and structure of rock slices,naked eye observation often requires researchers to master rich geological knowledge and identification experience.At the same time,the identification work is time-consuming and easy to be interfered by human factors.The purpose of this paper is to study an identification scheme based on computer technology to achieve the purpose of automatic identification of rock slices by computer.Under the tide of the big data era,the widespread popularization of geological data digitization and the rapid development of deep learning technology have provided abundant research materials and research methods for the experiments in this article.This paper proposes a multi-feature verification classification scheme for the sandstone rock slice image under the 10x orthogonal polarizer.The classification scheme is based on convolutional neural network and digital image processing technology.The main research contents of this paper are as follows:(1)Data collection arrangementIn view of the characteristics of the input data of the convolutional neural network,the image collection of this article needs to be sorted out.The number of various types of images in the data set is different.In order to solve the problem of image imbalance,oversampling and under-sampling are used for various types of images.At the same time,random rotation and scaling are added to enrich the data set type to balance the number of different types of final input data.Improve the generalization ability of the network model.(2)Selection and optimization of network modelThere are many types of convolutional neural network models.Aiming at the classification problem studied in this paper,by comparing the performance of the two types of network models VGGNet-16 and GoogLeNet-inceptionV3,the network model that is more suitable for this research is selected.By comparing the influence of learning rate on the results of the network model,the parameters that are more suitable for the data set in this paper are determined,thereby improving the recognition accuracy and running speed of the network model.(3)Feature enhancement and extractionIn order to further improve the recognition accuracy of the convolutional neural network model,this article uses dark channel defogging algorithm to enhance the image characteristics for some blurred images.At the same time,the image is standardized and normalized to improve the generalization ability and computing efficiency of the network.The statistical method is used to calculate the six texture descriptors of the image cooccurrence matrix,so as to achieve the effect of extracting image texture features.(4)Classification schemeBased on the sandstone triangle naming idea,a multi-network classification system for stepwise recognition is constructed.For the input rock slice images,multi-sampling is adopted to avoid the interference of the overall classification scheme due to the local factors of the image,and each image is combined with direct classification network features,step-by-step classification network features,and image texture features to score the highest value type Determine the recognition type of the final output.Using 1000 images as the test set to test the recognition effect of the classification scheme in this study,the final recognition accuracy rate was 88.7%.The results show that the classification scheme in this paper has a good effect on the identification of sandstone rock slices,and has a certain promotion and practical application value.