Convolutional Neural Network Based On Improved Support Vector Machine Research On Image Recognition Method

With the development of Internet technology and the geometric growth of image information,how to extract the effective information from a large amount of data quickly and accurately has become a difficult problem that people need to solve.The degree of image feature extraction determines the quality of real-time image recognition and recognition accuracy.Convolutional neural network is an identification method that has been developed in recent years and widely recognized by experts and scholars.The multi-layer network structure of convolutional neural networks can extract more abstract features from images and use more abundant image features for image recognition.Based on the study and analysis of existing image recognition methods and convolutional neural network algorithms,this paper introduces the direct support vector machine,twin support vector machine and ant colony algorithm into the convolutional neural network architecture.This paper further studies the application of convolutional neural network algorithm in the realization of image recognition in order to obtain better recognition of real-time or accuracy.For convolutional neural network based on support vector machine algorithm in the image recognition process,the problem of using SVM training and classification time is longer.A convolutional neural network based on direct support vector machine image recognition algorithm is proposed.CNN-DSVM first uses DSVM to perform inverse operations on nuclear matrices to reduce algorithm complexity and Uses DSVM as a classification function of convolutional neural network Algorithm.Convolutional neural network is used to extract the features of the training image sample set.Sets are used as the input of multiple DSVMs respectively to construct the CNN-DSVM classifier.The final experiment uses GTSRB database and Yale B database for simulation verification.Experimental results show that compared with CNN-SVM algorithm,CNN-DSVM reduces the image classification time by 11% and 16%,respectively,under the condition that the accuracy of image recognition remains basically unchanged.Aiming at the problem that the classification accuracy of CNN-SVM algorithm is not high in image recognition process,a convolutional neural networks based on ant colony optimization twin support vector Machines(CNN-ACO-TWSVM)is proposed.Firstly,the convolutional neural network is used to extract training image features.Secondly,the ant colony algorithm was used to optimize the penalty parameters of the twin support vector machines to construct multiple TWSVM models.The convolutional neural network is used to extract training image features as the input of TWSVM,and multiple CNN-ACO-TWSVM classifiers are constructed.Finally,the CASIA WebFace database and the dogs_vs_cats database were used for simulation verification.Experimental results show that the CNN-ACO-TWSVM algorithm improves the image classification accuracy by 8.93% and 9.93%,respectively,compared to the CNN-SVM algorithm.