Research Of Images Classification Based On Support Vector Machine And Semi-supervised Deep Belief Network Learning

Image classification is a hot research topic in the field of computer vision,with the growing diversity of cyber source,the image is larger and larger,more complicated,image classification technology is facing great challenges.In this paper,the related technologies of image classification are summarized,and the existing classification problems are summarized.In the light of the singleness and limitation of the small scale image classification technology in the past,and the existing problems of low precision and large scale image,this paper gives different solutions and expounds in detail.In order to explore the higher accuracy of image classification technology,this paper proposes a improved Daubechies wavelet based on fast PCA and SVM algorithm for image classification with the help of Daubechies wavelet's good local properties(DW-FPSVM).Firstly,facial features are extracted through the two-dimensional Daubechies wavelet decomposition.Secondly,reducing dimensions and demonizing are fulfilled by rapid PCA.Finally,by selecting appropriate kernel parameters,facial features classifications are proposed by SVM algorithm.This paper simulate the experiment with ORL face data sets,experimental results show that the face classification accuracy decreases with SVM kernel parameter increases and the improved Daubechies wavelet decomposition algorithm can effectively improve face recognition accuracy and stability.Besides,comparing to other algorithms in the training time and classification time,and the efficiency of the algorithm is proved.Aiming at the existing problems of low precision on the large-scale image classification,this paper proposes an adaboost algorithm in large-scale image classification based on deep feature learning.Using DBN as weak classifier to learn the sample image,the weights are adjusted by the error rate and the classification accuracy of each sample.Then the BP operator is used to readjust the sample weight and output the final error rate of each classifier.Finally,all weak classifiers are integrated into strong classifier and output the final classification results.This paper simulate the experiment with two kinds of data sets,they are MNIST and ETH-80.Comparing the classification results to other algorithms,show that the classification accuracy of this algorithm is higher than others.Realizing high precision large-scale image classification.