Hyperspectral Image Classification Method Based On Spatial Spectral Union

With the development of remote sensing technology,the resolution ratio of hyperspectral remote sensing images is getting higher and higher,and the information contained in the hyperspectral remote sensing images is more abundant,and the rich spectral information and spatial information contained in hyperspectral remote sensing images provide a new opportunity for the fine-tuning of the classification of earth objects.However,it also brings some problems that ultra-high resolution leads to a number of problems such as large amount of data,high redundancy,large number of bands and strong correlation,etc.If the high-spectral image is classified directly,it will not only consume a lot of time but also make a low classification accuracy.Therefore,if we can first denoise the hyperspectral image and then combine the rich spectral information and spatial information contained in the hyperspectral image,it can not only reduce the computation workload,but also improve the classification accuracy of the hyperspectral image classification algorithm.There are many classification methods for hyperspectral remote sensing images,but most classification algorithms focus on the application of spectral information of images,and do not take into account the continuity of the spatial distribution of the objects on earth,thus do not integrate spatial information into the classification algorithm of hyperspectral images,based on which the research contents of this paper are as follows:(1)Brief introduction has been made that traditional hyperspectral image classification algorithm is divided into two kinds.The first kind is the algorithm that only uses spectral information and the other kind is the algorithm that contains spatial information.Moreover,the evaluation index of the classification accuracy of hyperspectral image classification algorithm and the commonly used hyperspectral image data set have been introduced.(2)A hyperspectral image classification algorithm of spectral-spatial nearest neighbor(SSNN)is proposed.The algorithm's spectral-spatial information integration mainly utilizes the continuous characteristics of the ground material in spatial distribution,and the near neighbor space of the test sample point is first constructed,then the space near neighbor point in the near neighbor space that is not consistent with the test sample label is filtered,thus the interference of the heterogeneous point in the near neighbor space to the classification of test sample point is effectively solved,besides,Salt-and-pepper noise of images is filtered and improved.Secondly,a new distance measurement method is proposed that by introducing the regularization coefficient,calculating the distance between the training sample and the test sample near the neighbor space,the training sample label of minimum distance as the test sample label is selected.The overall classification accuracy of the algorithm on the Indian Pines and PaviaU hyperspectral data sets is 95.31% and 97.51%,respectively.(3)A hyperspectral image classification algorithm of spectral-spatial weighted nearest neighbor(SSWNN)is proposed.The algorithm first looks for the near neighbor space of the central pixel point and denoises the image.Secondly,the spatial information and spectral information of the picture element in the near neighbor space are fully integrated,and the weight of each near neighbor points is calculated according to the spectral similarity of the near neighbor point and the central picture element.Finally,the distance of the training sample near-neighbor space that has been denoised and weighted is resolved,with the label of the nearest training sample as the label of the central picture element.The overall classification accuracy of the SSWNN algorithm on the Indian Pines and PaviaU hyperspectral data sets is 96.75% and98.54%,respectively.In summary,classification algorithm based on hyperspectral image spectral information and spatial information fusion are mainly studied in this paper,the Salt-and-pepper noise is improved,and a new distance measurement method is proposed.Furthermore,the weighted fusion of the spatial spectrum was brought in,and the validity of the algorithm proposed in this paper through experiments on the hyperspectral image data set was verified.