Research On The Classification Algorithm Of Hyperspectral Remote Sensing Images Based On Spectral Sequence Examples

With the continuous popularization of hyperspectral imagers in recent years,hyperspectral remote sensing images have begun to be widely used in surface object recognition and classification,disaster and anomaly detection,urban evolution,national security and other important fields.Among these extensive fields,a large part of them can be summarized as the application of hyperspectral remote sensing image classification,so the classification of hyperspectral images is a siginificant research topic.At the same time,the classification of hyperspectral images has become a difficult problem due to the characteristics of hyperspectral data itself,such as large dimensions,few samples,homospectral and heterospectral.Therefore,this paper starts from the analysis of spectral information,and gradually discusses the classification of hyperspectral remote sensing images.In view of the current problems,this paper proposes a spatial spectral feature extraction classification model with spectral information as the main body on the basis of relevant research:The followings are the finished works:1)A LSTM hyperspectral image classification method combined with guided filtering is proposed.After analyzing and confirming the feasibility of spectral information as the basis of classification,the proposed method firstly makes use of guided filtering to process the original image data and integrates spatial information into spectral information;Then,LSTM is used to extract features of the newly obtained spectral sequence,so as to obtain features with both spatial information and spectral information;Finally,the feature is fed into the classifier for classification.In order to verify the validity and feasibility of the classification method proposed in this paper,three commonly used data sets are respectively verified.The experimental results show that the proposed model can achieve high classification performance after training,and has strong competitiveness in the case of sufficient and insufficient samples.Therefore,the proposed method can be proved to be effective in the classification of hyperspectral remote sensing images.2)Since the spatial information in 1)is fused into spectral information by guided filtering technology and then the new spectral sequence is modeled and analyzed,which destroys the original state of spectral information and is not an end-to-end overall classification algorithm.Therefore,in view of the hidden dangers and shortcomings of the method in 1),a Shared weight conv-lstm feature extraction classification algorithm is proposed.Firstly,the concentric data blocks are obtained by modeling the original image;Then,the data block is extracted using the conv-lstm of Shared weights to obtain the feature block;and then,use max-pooling to conduct feature selection and dimension reduction for feature blocks;Finally,classifier is used to classify features.Experimental results show that the proposed algorithm can effectively extract and classify features,and is still competitive in the case of sufficient samples,insufficient samples,complex sample composition and simple sample composition.Therefore,it can be proved that the proposed method is practical and feasible for different characteristics of hyperspectral image data.