Semi-supervised Image Classification Based On Improved Ladder Network

Semi-supervised classification is an important research direction in the field of image classification.It trains the classifier with a small amount of labeled data samples and assists the training process with a large amount of unlabeled data,so as to improve the performance of the classifier.With the continuous development of deep learning,semi-supervised classification technology has been improved qualitatively,and it has become an important research direction to deal with big data classification.How to make full use of unlabeled data is the key to improve the classification performance of semi-supervised models.Meanwhile,the performance of semi-supervised models depends on the influence of network structure,loss function and classifier selection to a large extent.In order to solve the problem that the semi-supervised deep generation network model cannot fully apply the unlabeled data,on the basis of deep neural network frameworks such as ladder network,constraints such as large margin sofmax loss function and virtual adversarial training are introduced.The semi-supervised classification model based on ladder network and modified tri-training algorithm and the semi-supervised classification model based on mix?up data augmentation and virtual adversarial training are proposed.The main research results are as follows:1.For semi-supervised deep generation models,it is easy to cause over-fitting problems,a semi-supervised virtual adversarial training model based on improved ladder network is proposed.The model trains the classifier with a combination of mix?up data augmentation and virtual adversarial training based on the ladder network.First,use mix?up to enhance the training data to obtain new extended data to solve the problem of fewer labeled samples in the semi-supervised classification model.Next,the virtual adversarial noise is applied to the ladder network,and the generalization ability of the model is improved by constructing smooth regularization constraints.Finally,the model adjusts the parameters by combining the classification loss of labeled data,the reconstruction loss of unlabeled data and the virtual adversarial loss,and trains to obtain a classifier.The model is tested on MNIST,SVHN and CIFAR10 respectively.In comparison with other semi-supervised deep generation models,test results show that the model proposed in this paper has good application effect to improve the generalization and has obtained better classification accuracy over existing semi-supervised.2.In order to improve the classification performance of semi-supervised deep generation models,a new model based on ladder network and improved tri-training algorithm is proposed.This model adds three classifiers to the highest layer of the noisy coding layer of the ladder network,and improves the image classification performance by combining the improved tri-training.First,the labeled data is divided into three parts by using a class-based sampling method.The model adjusts the parameters by combining the labeled error of labeled data and the re-construction error of unlabeled data,and is trained to get three large-margin softmax classifiers.Next,the improved tri-training algorithm is developed to add pseudo-labels to the unlabeled data,and different weights is as-signed to the new labeled data to expand the training set.Then,the expanded training set is applied to update the model.After the above practice,a weighted voting is performed on the classifier to obtain the classification result.The characteristics of the ladder network obtained by this model have better low-dimensional manifold representations,which can effectively avoid classification error caused by uneven distribution of sample data and enhance generalization ability.The model is tested on the MNIST,SVHN and CIFAR10 respectively.In comparison with other semi-supervised deep generation models,test results show that the model proposed in this paper has obtained state-of-the-art classification accuracy over existing semi-supervised learning methods.