The Classification Of Hyperspectral Image Based On Spatial Filtering And Generative Adversarial Network

With the development of hyperspectral remote sensing,the resolution of hyperspectral data is becoming higher and higher.It plays more and more important roles in various fields of social life.The hyperspectral data include many spectral and spatial information.With the abundant spectral and spatial information in the hyperspectral data,the pixel-wise classification of hyperspectral data can be accomplished well.In this paper,some new classification methods for hyperspectral data have been proposed,and the purpose of the paper is to improve the classification result for the hyperspectral data.The paper mainly includes the following aspects:With the spatial resolution of hyperspectral data becoming more and more higher,the spatial information becomes more and more valuable for the processing of hyperspectral data,thus the spatial information has been used in this paper.The paper proposed a classification method for hyperspectral data which is based on spatial filtering,the spatial filtering decreases the classification time and increases the classification accuracy.And the paper analyzed several existing classical classification algorithms which incorporate the spatial and spectral information systematically.The experimental results showed the efficiency and effectiveness of the filtering method.The generative adversarial network is composed of two models: a generative model and a discriminative model.This paper applies the generative adversarial network to the classification for hyperspectral data.And the generative adversarial network has been changed into a classification network instead of a generative network.Specifically,the discriminative model has been used as a classifier and the loss function has been changed so that it becomes a semi supervised classification model for hyperspectral data.The experimental results demonstrated it performs well in the classification of hyperspectral data.