Research On Hyperspectral Image Classification Method Based On Spectral And Spatial Joint Information

Hyperspectral image contains a lot of spectral and spatial information.It can better identify and classify target objects and play an important role in resource survey,agricultural pest detection,biomedicine,and other fields.In the task of hyperspectral image classification,how to make full use of the spectral and spatial joint information of the hyperspectral image is the key to improving the classification accuracy.At present,the hyperspectral image classification method based on deep learning has some problems,such as a large amount of model operation,too single convolution scale,insufficient utilization of the relationship information between the context of spectral sequence,the spatial information of the image is not fully utilized,which will affect the final classification result.In view of the above problems,according to the characteristics of the hyperspectral image and the research status of hyperspectral image classification,under the framework of deep learning,this paper proposes two hyperspectral image classification methods based on spectral and spatial joint information to realize the effective recognition and classification of target objects.The main research contents are as follows:（1）This paper proposes a hyperspectral image classification method based on attention mechanism and multi-scale hybrid CNN.This method is based on the spectral and spatial joint information of the hyperspectral image.Firstly,two-layer threedimensional convolution is used to extract the spectral-spatial features;Then,twodimensional convolution operations of different scales are used to further extract spectral-spatial features,which not only reduces the computational complexity of deep3D-CNN but also extracts multi-scale spectral-spatial features;Secondly,the method also adds channels and spatial attention modules to the model,which can strengthen and obtain more effective spectral-spatial features;Then,the spectral-spatial features of different scales are fused to obtain a more abstract multi-scale spectral-spatial joint features;Finally,the joint features are input into the Softmax classifier for classification.（2）In order to better learn the context information between spectral bands and make full use of the spatial information of the hyperspectral image,a hyperspectral image classification method based on Bi-LSTM and deep and shallow spatial features is proposed in this paper.Firstly,Bi-LSTM is used to learn the context information between spectral sequences,so that more effective spectral features can be extracted.Then,in the process of spatial feature extraction,considering the possible loss of important details in the convolution and pooling operation of the 2D-CNN model,this method combines the spatial features extracted by each layered network of the 2D-CNN model,so as to obtain an abstract deep and shallow spatial features,and then fuse the obtained spectral features with deep and shallow spatial features to obtain a more robust spectral-spatial joint features;Finally,the joint features are input into the Softmax classifier for classification.Based on the hyperspectral image data sets of Indian Pines,Pavia University,and Salinas,the two proposed methods are compared with other classification methods.Experimental results show that the proposed method effectively improves the classification accuracy of the hyperspectral image and obtains better classification results.