Hyperspectral Image Classification Method Based on Multi-class Generative Adversarial Network

Abstract

The invention discloses a hyperspectral image classification method based on a multi-class generative confrontation network. The steps are as follows: (1) input hyperspectral images; (2) obtain sample sets; (3) generate training samples and test samples; (4) ) Build a multi-class generative confrontation network; (5) use the generator to generate samples; (6) use the discriminator to classify the training samples and the generated samples; (7) build the loss function of the generator and the discriminator; (8) alternate Train generator and discriminator; (9) Classify hyperspectral images. The present invention utilizes the built multi-class generative confrontation network to extract the spatial features of the pixel neighborhood, and simultaneously generates samples to increase the number of samples, enhances the feature extraction capability of the network, alleviates the problem of network over-fitting, and improves the hyperspectral image quality. classification accuracy.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to a hyperspectral image classification method based on multi-class generative adversarial networks (multi-class generative adversarial networks) in the technical field of image classification. The invention can be used to classify ground objects of hyperspectral images. Background technique [0002] The improvement of the spectral resolution of hyperspectral images provides more abundant information for classification, but also brings great challenges. Traditional methods currently used in hyperspectral image classification include support vector machines, decision trees, etc., and methods based on deep learning include stacked autoencoders, convolutional neural networks, etc. Deep learning requires a large amount of labeled data as training samples, and it is difficult to collect enough labeled data in hyperspectral images. Therefore, in the classification of hyper...

AI Technical Summary

Problems solved by technology

Although this method can reduce the influence of radiation error and geometric error, the disadvantage of this method is that only the spectral features of the pixel are extracted, and the spatial features of the neighborhood of the pixel are not extracted, resulting in low classification accuracy.

Although this method retains the nonlinear information of the samples, the disadvantage of this method is that the number of samples is too small relative to the number of network parameters, which leads to over-fitting of the network and low classification accuracy.

Images