Classification And Implementation Of Rock Slice Image Based On Deep Learning

In the field of geology,the recognization of rock species is of great significance for geological exploration.Compared to the external representation of the rock,the finer rock slices have more intuitive understanding of the internal fine structure to classification.The classification algorithm of rock slice images,which is based on deep learning,helps to improve the objectivity of image classification,reduce labor costs and improve classification accuracy.Therefore,this paper proposes a classification method of rock slice images based on multifusion miniature convolutional neural network by analyzing the color and extinction characteristics of rock slice images.The main research of this paper are as follows:1.In order to improve the overall quality of rock slice images and enhance the details of the images,an image super-score method based on asymmetric generative adversarial networks is proposed.Aiming the problem that the existing image super-resolution reconstruction methods have weak ability to restore high-frequency information of images and cannot capture the key features of each channel,the high-order degradation method is introduced,which simulate the real-world image degradation process,and a low-resolution rock slice image dataset is obtained by this method;Secondly,an asymmetric attention residual dense block is constructed to improve the effective fusion of image shallow features and deep features,and a multi-scale discriminator is constructed to judge the authenticity of images on three scales with the help of the discriminator.Experimental results show that compared with existing superresolution methods,the reconstructed images in this paper contain more high-frequency information and detailed features.2.Geological exploration sites are mostly in the wild.In order to facilitate the deployment of existing classification models into portable mobile devices,lightweight technology is introduced that reduce the existing complex models.But it also brings the problem of reducing model’s classification accuracy while improving its efficiency.Therefore,this paper proposes a rock slice image classification model based on multi-fusion miniature convolutional neural network to reduce the amount of model parameters and ensure high classification accuracy.This method improves the classification accuracy of multi-source fusion images through image fusion strategy,image classification method and result fusion mechanism.The experimental results show that the image classification accuracy is improved with using lightweight images,which proves that the method in this paper is effective.To sum up,the method proposed in this paper has small parameters and relatively low computational complexity,which can effectively improve the accuracy of rock slice classification and provide technical support for mobile portable equipment for geological exploration.