Image Classification Based On Deep Convolutional Generative Adversarial Networks

With the development of human society,a large amount of image data has appeared in people's daily lives.People need to efficiently process these image data to serve human beings.Image classification is an image treatment technology that divides images into different categories according to the different features in the image information.Nowadays,image classification is mostly implemented using convolutional neural networks.Traditional convolutional neural networks generally require a large labeled data set,but in actial life such data sets are relatively difficult to obtain.Therefore,it is necessary to use only a small amount of labeled data to train a convolutional neural network with good performance.First,this article briefly introduces the research background and actuality of convolutional neural networks.Then,for the problem that the traditional convolutional neural network can not use unlabeled data,A semi-supervised learning model based on Convolutional Neural Networks and Generative Adversarial Networks is proposed.And complete the following work:(1)Starting from the artificial neural network,the theoretical basis of the artificial neural network and the basic process of training the neural network with the back propagation algorithm are described.After that,the characteristics and advantages of the convolutional neural network are introduced in detail.The classic network LeNET-5 is used as an example.(2)In order to use a large amount of unlabeled data in the data set,a semi-supervised learning model is constructed based on the convolutional neural network and the generated adversarial network.The basic structure of the model and the training process are introduced in detail.And explained the principle of semi-supervised learning.(3)The model proposed in this paper implements semi-supervised training on the MNIST dataset,and CIFAR-10 dataset,SVHN dataset.And the experimental results are compared with the traditional semi-supervised model.The experimental results prove that the model examined in this paper has higher classification accuracy.