Hyperspectral Image Classification Based On Deep Learning And Spectra-spacial Combination

Hyperspectral image technology has transformed from a scarce research effort into a wide range of community available products for users over the past four decades.Hyperspectral images contain abundant spectral curves of ground objects.The traditional classification method only considers the spectral information of the hyperspectral image and ignores the spatial information,resulting in a poor classification result.With the development of imaging technology,the spatial resolution of hyperspectral image gradually increases,the spectralspatial-combined method becomes more and more popular among researchers.However,the methods for extracting spatial information are very different from each other,and the labeled samples are scarce.How to extract the spatial information of hyperspectral images effectively and augment dataset becomes the key of research.In recent years,with the development of artificial intelligence,deep learning technology has gradually entered into people’s field of vision.Deep learning firstly came out in image recognition,but in a few short years,it spread to all areas of machine learning.Today,deep learning has a very good performance in many machine learning areas.This paper takes use of the advantages of deep learning in feature extraction to apply deep learning to the task of classification of hyperspectral images.The main contributions are as follows:1)Aiming at the difficulty in training deep networks and low efficiency in feature extracting,a hyperspectral image classification method based on Clique Net is proposed.The recurrent feedback is able to bring higher level visual information back to refine low level filters and achieve spatial attention.The method first uses PCA to reduce the original hyperspectral data,and then send three pcs to the Clique Net to extract features The experimental results show that the hyperspectral classification method based on Clique Net is highly efficient.2)For the problem that the traditional spatial feature extracting method is single and the labeled data is very rare,a hyperspectral image classification method based on feature fusion and generative adversarial network is proposed.This method fused three spatial features and then send it into discriminant network with the fake samples produced by generative networks to classify.This method does not only augment the data,but also extract efficient spatial features.The experimental results show that the method has a very good performance in the hyperspectral classification work.3)Inspired by the observation that CNN learned features are naturally coupled with intraclass variation and the semantic difference.So it is hard to judge the intra-class variation and the semantic difference when the norm is very high.We proposed a hyperspectral image classification method based on decoupled network which models the intra-class variation and semantic difference independently.We first reparametrize the inner product to a decoupled form and then generalize it to the decoupled convolution operator.The method is able to extract context features more efficiently,and the network converge very fast.The experimental results show that the decoupled network also performs well in the task of classification of hyperspectral remote sensing images.