A method and system for image diversity enhancement based on generative adversarial network

Abstract

The invention discloses an image diversity enhancement method and system based on a generative confrontation network, which belongs to the field of machine learning technology. In order to solve the problems of small sample learning, sample expansion and balance of arbitrary image data sets, etc., the technical solution of the invention includes a generator module, discriminator module and cluster visualization module. The generator module uses noise to generate new adversarial sample images; the discriminator module compares the original image data set with the adversarial samples for "true and fake"; the cluster visualization module calculates the population size of the generated adversarial samples. This invention innovatively combines the "compression coding function" with the "Simpson diversity index", and proposes a classification-oriented diversity maximization loss function, which can effectively improve the population while ensuring the fidelity of the generated adversarial sample images The diversity of the inner sample.

Description

technical field [0001] The invention belongs to the technical field of machine learning, and in particular relates to an image diversity enhancement method and system based on a generative confrontation network. Background technique [0002] Generative Adversarial Networks (GAN, Generative Adversarial Networks) is a deep learning model proposed by Goodfellow et al. in 2014, and it is one of the popular models in the deep learning field in recent years. The original generative adversarial network model framework has two modules: Generative Model and Discriminative Model. The two modules achieve the effect the user wants through the "game" between them. In generative adversarial network instances, deep neural networks are often used as the generation module and the discriminant module. A "well-trained" generative adversarial network has a very high demand for a suitable training method, otherwise the training results may be unsatisfactory due to the unsupervised nature of th...

AI Technical Summary

Problems solved by technology

However, there will be a problem of mode collapse when improving the generative confrontation network through the existing technology.

The generated images are the same, and the diversity of the images has not been enhanced.

Images