The Study Of Hyperspectral Image Classification Method Based On Parameterization Of Object Spectral Profile Shape

The spectral morphological differences feature, however which has not been taken advantage in the existing remote sensing image classification method, is the important basis to identify objects. On the basis of in-depth understanding of the mechanism of remote sensing imaging and the theory of object spectrum, the method of typical object spectrum shape parameterized in the the study area was implemented firstly through descriping curve shape based on morpheme vector; and then the remote sensing image classification procedures and methods based on the shape differences of object spectrum were designed and completed in Matlab, taking the classification criteria criteria that different objects have distinct spectral curve shape and the matching algorithm with wildcards for spectral morphology. The method proposed in this article maximized the use of the waveform (peaks and troughs, up, down, etc.) information by emphasizing the "shape" similarity characteristics of the object spectrum curve, which broke the problems that previous study of remote sensing image classification mostly concerned about the spectral "values" similarity characteristics and were influenced largely by the change within the class, compared with other classification methods suitable for hyperspectral remote sensing image. In addition, the accuracy evaluation results of the Kappa coefficient showed that the overall accuracy and Kappa coefficient of the method used in this paper were higher than the SVM classification method. The research achievement will not only deepen the theory and practice of hyperspectral remote sensing image classification, but also be suitable for vegetation remote sensing identification and quantified land-cover classification with obvious characteristics of seasonal variation.