A hyperspectral image classification method based on a cooperative generation antagonistic network and a space spectrum combination

Abstract

The invention discloses a hyperspectral image classification method based on a cooperative generation antagonistic network and a space spectrum combination. The method comprises the following steps: inputting a hyperspectral image; obtaining a sample set; generating training samples and test samples; building a multi-scale discriminator; constructing cooperative relationship; building a cooperative generation countermeasure network; the initial value of the training sample being generated by the multi-scale discriminator; the generator generating samples; a multi-scale discriminator perfromingclassification; constructing the loss function of generator and multiscale discriminator; alternatively training the generator and the multi-scale discriminator; hyperspectral images being classified. The invention utilizes the established cooperation to generate the antagonistic network, extracts the spatial spectrum combined characteristics of pixels, generates more realistic samples, increasesthe number of samples, alleviates the problems of over-fitting the network and slow convergence speed of the network, and improves the accuracy of the hyperspectral image classification.

Description

technical field [0001] The present invention belongs to the technical field of image processing, and further relates to a hyperspectral image classification method in the technical field of image classification through collaborative generative confrontation network and space-spectrum union. The invention can detect and recognize underwater obstacles and land tanks and warships in the collected hyperspectral images, and can analyze the types of crops in the hyperspectral images. Background technique [0002] Hyperspectral images can obtain approximately continuous spectral information of target ground objects in a large number of bands such as ultraviolet, visible light, near-infrared and mid-infrared, and describe the spatial distribution relationship of ground objects in the form of images, so as to establish "map-spectrum integration" data , can realize accurate identification and detail extraction of ground objects, and provide favorable conditions for understanding the o...

AI Technical Summary

Problems solved by technology

Although this method does not need to solve the complex global energy optimization problem, it can extract the non-local spatial information of the hyperspectral image to optimize the classification. For the problem of large and small differences in the spatial characteristics of different pixels, only extracting the spatial characteristics of the hyperspectral image will lead to a large number of misclassifications, resulting in low classification accuracy

Although this method makes full use of the characteristics of "integration of space and spectrum" and "double high resolution" of hyperspectral data, the disadvantage of this method is that when using convolutional neural network training, the network parameters are many The network training time is long, and the number of samples is too small relative to the number of network parameters, resulting in over-fitting of the network, resulting in low classification accuracy

Images