Research On The Identification And Classification Of Mineral Composition Of Igneous Rocks Based On Convolutional Neural Network

Igneous rock is one of the main rocks that constitute the earth crust and contains important mineral reserves.Therefore,it is of great significance to analyze its mineral composition.However,the traditional rock mineral analysis method is tedious,complicated,time-consuming and inefficient.In this paper,based on the principle of CT imaging,combined with digital image processing technology,unsupervised feature extraction and image segmentation algorithm,preprocessing,feature extraction,segmentation and other operations are carried out on igneous rock CT image to realize the identification and classification of mineral components.In view of the interference factors such as noise,blur and background interference in the CT imaging process,the CT imaging of igneous rocks is pre-processed,including image cropping,image enhancement and filtering.The processed image has higher definition and more obvious outline,which is more suitable for experimental research.In order to better identify and classify the mineral composition of igneous rocks,this paper proposes an unsupervised feature extraction model based on a symmetrical fully convolutional neural network,which is improved on the basis of traditional autoencoder and convolutional neural networks.The entire model is composed of convolutional layers and deconvolutional layers,which overcomes the traditional autoencoder only accepts the input in the form of vectors and the large number of training parameters of the fully connected structure leads to large memory occupation,slow calculation speed and overfitting.Removal of the pooling layer of the traditional convolutional neural network better maintains the regional details of the image,which is conducive to the reconstruction of the image.A deconvolution layer is introduced to enable the model to reconstruct the image by using sparse higher-order features.The error between the reconstructed image and the original image is small,which proves that the convolutional layer has extracted the main features of the original image.The trained model can be used to extract the features of igneous rock CT imaging.The process of model training and feature extraction are unsupervised,which solves the problem that the supervised feature extraction method is not applicable.Finally,through the establishment of the igneous rock mineral composition identification and classification model,the content of each mineral composition is analyzed,and then the name of the identified and classified mineral composition canbe confirmed according to the igneous rock mineral composition content law.