Research On Hyperspectral Image Classification Method Based On Spatial Spectral Combination

Hyperspectral images are acquired by hyperspectral remote sensors through continuous remote sensing imaging of the same feature area in hundreds of continuous narrow spectral bands.Therefore,the image contains rich spectral information and spatial information,which is used in the field of feature recognition and classification.It has huge advantages and plays an increasingly important role in the fields of mineral exploration,precision agriculture,disaster prevention and control,environmental monitoring and military detectives.Hyperspectral image classification firstly extracts the effective information that can be used for classification from the spectral information and spatial information of the hyperspectral image through the feature extraction algorithm,and then uses the classifier to classify all pixels in the image into different feature categories.Due to the data characteristics of hyperspectral images,the dimensional disaster caused by high data dimensions,the information redundancy caused by the high correlation between bands,the same spectrum of foreign matter,the same spectrum of foreign matter,etc.,all bring to the feature extraction and classification of hyperspectral images.In view of the enormous challenges,it is essential to develop effective and robust classification methods.On the basis of summarizing the current status of hyperspectral image classification,this thesis proposes two hyperspectral image classification methods.The main work of this article is as follows:(1)A deep learning classification algorithm for hyperspectral images based on two-channel variational autoencoder is proposed.By building a one-dimensional conditions variation since the encoder feature extraction framework and the conditions of two-dimensional circular channels variational feature extraction from the encoder framework respectively to extract the spectral features and spatial features of hyperspectral image,and then the spectrum feature vector and spatial feature vector superposition formed empty spectrum joint feature vector,and finally the joint feature into Softmax classifier to classify.In order to solve the problem that small sample data is not good for training,image enhancement technology is used to expand the training sample.The experimental results show that the proposed classification method based on spatial-spectral feature shows better classification effect.(2)A classification method based on the combination of Gabor filter and convolutional neural network is proposed,which effectively combines Gabor filter to extract the edge and texture information of hyperspectral image.Two dimensional convolutional neural network model and Gabor filter are used to process the original hyperspectral image data to extract deep spatial information.At the same time,one-dimensional convolutional neural network model and three-dimensional Gabor filter are used Combined with deep spectrum feature extraction.Finally,the two CNN model output layers are connected to the same fully connected layer to achieve effective fusion of the two features,and the combined features after the fusion are used to complete the classification..Experiments show that the proposed method achieves good classification results.