Image Classification Based On Multiple Kernel Domain Adaptive Sparse Representation And Deep Convolutional Neural Network

With the advent of the era of big data,massive amounts of image data need to be processed.In order to mine the wealth of information available in these image data,the demand for image processing algorithms is also increasing.As a basic research in the field of image processing,image classification can promote the overall development of computer vision and image processing,has a wide range of applications and has become the research focus of researchers in the field of image processing.This paper focuses on two types of image classification algorithms.Aiming at the problem that small amounts of data come from different domains,a method based on multi-core domain adaptive sparse representation classifier is designed to enable the classifier to obtain information common to the source domain and the target domain,thereby expanding the ability of domain adaptation.Aiming at the situation where large sample size data is not optimized for domain-specific data,we focus on ground-based cloud image classification and design a ground-based cloud image classifier based on a deep convolutional neural network to solve the problem of decreasing ground-based cloud image classification accuracy with mixed resolution.The main work of this article is as follows:(1)In the multi-core sparse representation classifier,the domain adaptive method is introduced to construct the multiple kernel domain adaptive sparse representation classifier,which enables the multiple kernel sparse representation classifier to learn data in different domains together and based on the decision criteria of the multiple kernel domain adaptive sparse representation classifier,and then a multiple kernel domain adaptive discrimination projection method is proposed.By jointly learning data projections from two domains and a universal discriminative dictionary,a better representation of data from different domains in the projection space is obtained.Then an alternative iterative algorithm is proposed to make the above algorithm realized.And the superiority of the multiple kernel domain adaptive classification algorithm is proved by experiments.(2)Proposing a mixed-resolution ground-based cloud classification method based on deep convolutional neural network,and set up a mixed-resolution ground-based cloud image data set(MGCD).By introducing batch normalization and spatial pyramid pooling methods to the VGG network,the problem of classification accuracy degradation caused by image deformation of different resolutions in the preprocessing of ground-based cloud is solved,and the convergence of the network model is accelerated and the model generalization ability is improved.The method proposed in this paper achieves an average test accuracy of 95% in the mixed data set,which exceeds the current latest ground-based cloud map classification algorithms.