Zero-sample image classification method based on variational self-coding adversarial network

Abstract

The invention discloses a zero-sample image classification method based on a variational self-coding adversarial network. Two variational auto-encoders of a visual mode and a semantic mode are constructed, and visual features and semantic features are respectively and correspondingly used as inputs of the two variational auto-encoders; pseudo visual features and semantic features are generated, finally, the true visual features and the generated semantic features are input into a discriminator, and the adversarial process is completed through a metric learning method. Then starting to train asoftmax classifier, inputting the visual features of the non-visible class images into a variational auto-encoder of a visual mode, and training the classifier by utilizing the generated pseudo visualfeatures and corresponding labels; during testing, the real visual features of the non-visible class samples are input into the classifier for classification, and a zero-sample image classification task is realized. The method can achieve the classification task in a more real scene, facilitates the promotion of the application of zero-sample learning to the production and living reality, and accelerates the practical development of a deep learning algorithm.

Description

technical field [0001] The invention relates to an image classification method. In particular, it relates to a zero-shot image classification method based on variational autoencoder adversarial networks. Background technique [0002] For a long time, machine learning has received extensive attention in fields such as natural language processing, computer vision, and speech recognition. In recent years, in the field of computer vision, the performance of image classification tasks has been continuously improved, the application scenarios have been continuously extended, and the requirements for classification technology have become more and more stringent. With the development of deep convolutional neural networks, machine learning has reached a new level of development. Supervised learning is an important method of machine learning. In solving image classification problems, the performance of supervised learning can be obtained through a large number of manually marked imag...

AI Technical Summary

Problems solved by technology

However, due to the introduction of the variational lower bound of VAE, the generated visual features are easy to be distorted.

Images