Research On Classification For Hyperspectral Remote Sensing Based On Spatial Information

In the field of hyperspectral image classification,there are many ways to use spatial information.The fusion of the spatial features and the use of super-pixel information to optimize the results can effectively integrate the spatial information into the classification steps.Although these methods are effective,there are some shortcomings.The method of spatial spectrum feature fusion usually only integrates a spatial feature and a spectral feature into the classifier.However,the spatial information contains texture information,mathematical morphology information and neighborhood information,so a spatial feature can not fully describe the hyperspectral image's spatial information.The method of using the superpixel to optimize the classification result is to process the classification result after the classification,and the pixels in the super pixel region with the same number of tags are not able to effectively refer to the superpixel information,and the utilization rate of the information is not high.In this paper,we make a deep research on the utilization of many kinds of spatial information in hyperspectral classification.The research mainly includes the following aspects:Aiming at the problem that spatial information of hyperspectral image is not fully represented by a single spatial feature,this paper proposes a hyperspectral image classification based on adaptive multi-feature combination.Based on the differential morphological features,the algorithm finds the best one of the differential morphological features of different dimensions adaptively as the basic feature of fusion.Then,Gabor features and spectral features are fused to the optimal difference morphological features and sent to SVM classifier.The obtained results are compared with the classification results of the optimal differential morphological characteristics,and the optimal result is taken as the output.Finally,the SLICO spatial information is used to optimize the classification results.This method makes full use of the spatial information of hyperspectral image and the obtained features describe the hyperspectral image from many aspects.The adaptive selection of the optimal benchmark features ensures the lower bound of the algorithm accuracy and has good adaptability to hyperspectral images of different sizes and different contents.Aiming at the problem of low utilization rate of traditional superpixels information,this paper presents a method of "smooth moving",which is used to process the features before classifying.Superpixel information is added to each pixel feature vector of hyperspectral image.This method fully considers the neighborhood information of each pixel and has a certain noise reduction effect on the features before "smooth shift",which can effectively reduce the classification error and has better effect than post-processing.In this paper,this method is applied to the classification of hyperspectral images by adaptive multi-feature combination,which effectively improves the classification effect.