Zero sample image classification method based on generative adversarial network

Abstract

The invention discloses a zero sample image classification method based on a generative adversarial network, and belongs to the technical field of deep learning. The method comprises: acquiring an image data set; preprocessing the image data set to obtain a training set and a test set; constructing a core neural network, and inputting the training set into the core neural network to obtain image features and visual image features, wherein the core neural network comprises a convolutional neural network, a generative adversarial network, a reconstruction network and an attention network; calculating a loss function of the core neural network, and adjusting parameters of the core neural network; jointly training the picture image features and the visual image features to obtain a classifier; and inputting the test set into a classifier for classification. Compared with the prior art, the reconstruction network is added into the generative adversarial network, so that the visual image features generated by the generative adversarial network are more diversified; and the attention network is introduced into the generative adversarial network, so that interference information in visual image features is reduced, and the classification accuracy is improved.

Description

technical field [0001] The invention relates to a zero-sample image classification method based on a generative confrontation network, which belongs to the field of deep learning. Background technique [0002] With the development of deep learning in recent years, deep learning has also made breakthroughs in the field of natural image recognition, such as image recognition and classification, image text description, and image segmentation. The performance of object recognition and classification is particularly outstanding. However, during the development of the algorithm, shortcomings such as poor generalization ability and large training data are gradually exposed. However, the traditional method requires a large number of labels for image classification, and too large training data makes manual labeling difficult, so the traditional method cannot classify it. [0003] Larochelle et al. proposed the concept of zero-sample learning in 2008. With the great interest in zer...

AI Technical Summary

Problems solved by technology

However, the classification accuracy of the original zero-sample image classification method is low, and the generation confrontation network is prone to the problem of mode collapse. Only through the mutual confrontation between the generator and the discriminator will make the final generated image features tend to be single.

Images