Studies on image recognition based on VAE and AAE

Image recognition is applied to all aspects of life, such as brush face authentication and defect detection. Traditional convolutional neural network trains network through a large number of data sets. However, in practice, it takes a lot of time and effort to collect training data. In the case of less training data and larger verification data set, the accuracy is relatively low. Therefore, this paper proposes two methods this method can improve the accuracy and loss function fluctuation of the case with less training data and more verification sets. The lenet5 convolutional network structure is used as the main body to form a new convolutional neural network by combining variational self coding and counter self coding. The experimental results show that in the cifar-10 data set and cifar-100 data set, the combination of variational self coding and combined counter self coding can increase by up to 4 percentage points compared with the traditional convolutional neural network, which is effective in the case of insufficient training data.