Research On Hyperspectral Remote Sensing Image Recognition Based On Spatial And Spectral Characteristics

Traditional remote sensing image classification can only categorize objects in a general way.But with the development of spectral imaging technology,this simple classification can no longer meet people’s demand for automatic recognition of ground objects.Therefore,more attention has been paid to the fine classification of hyperspectral remote sensing images.In view of the characteristics of hyperspectral remote sensing images,a recognition method of hyperspectral remote sensing images based on multi-feature fusion classification and spatial correction of image segmentation is proposed in this paper.Firstly,the spatial and spectral features of hyperspectral remote sensing images are extracted and fused linearly.Then,SVM is used for preliminary classification.Finally,the preliminary classification results are corrected by using spatial features.The main work of this paper is as follows:(1)For hyperspectral remote sensing images,the use of spectral features alone or spatial features alone can not make full use of its information.In this paper,a multi-feature fusion method for hyperspectral remote sensing image recognition is proposed.Gabor feature,morphological difference feature and spectral feature are fused and classified by support vector machine.The three features extract the spatial texture information,light and shade information and spectral information of hyperspectral remote sensing image,which makes the method make full use of the huge amount of information of hyperspectral data and avoid redundancy between hyperspectral data after feature extraction.(2)Because it is difficult for spectral imager to avoid the influence of watermarking and noise when imaging,hyperspectral remote sensing images will have a lot of salt and pepper noise points and debris misclassification after preliminary supervised classification.In view of this error,this paper proposes a spatial feature correction method based on hyperspectral image segmentation.Firstly,the hyperspectral image is divided into small fragments by SLIC super-pixel segmentation,and then the segmentation results of hyperspectral remote sensing image are obtained by region merging of SLIC super-pixel segmentation results using fast nearest neighbor region merging method.The preliminary classification results are voted by the segmentation results to remove salt and pepper noise points.In order to ensure the validity of the segmentation results as much as possible,Before segmentation,spectral features are used to select the bands of hyperspectral data,which makes the dimension selected the most adaptable to the segmentation method.