Research On Scale Effect Of Remote Sensing Classification In Terms Of Spectral Variation

With the development of remote sensing images acquisition,processing and modeling,the application of remote sensing is constantly expanding.There is a urgent need for the matching of the classification accuracy and images resolution on various scales.As the main source of spatial information,the formation of the inherent uncertainty of images classification is the phenomenon of "spectral variation" and "mixed pixels".In previous studies,scholars mostly studied the scale characteristics of the net effects of “mixed pixels” and “spectral variation” or simply considered the spatial scale characteristics of uncertainties caused by mixed pixels.However,the study on the scale effect of classification based on spectral variation is quite rare.Conventionally,based on remote sensing images or thematic classification maps,the separability,variance,and measurement accuracy(confusion matrix)of spectral features varied bv resolution are used to study on the uncertainty extraction of remote sensing classification.However,the effect of each material texture parameter on inherent uncertainty of images classification is ignored.Therefore,a method was developed in this study based on controllable material texture simulation data for obtaining the ground model simulation data on different scales.The scale effect model of the classification accuarcy was established by taking into account the spectral effect variation.Finally,the conclusions were verified by the consistency between the simulation image and the real in Heihe City,Heilongjiang Province.The results showed that:(1)Only considering the phenomenon on different features obtaining the same spectrum of the spatial variation,the classification errors(omission error and commission error)gradually decreased with increasing images resolution.(2)The spectral variation period,the average intensity of the spectral variation,the spectral variation intensity,and the spatial shape of the spectral variation together influenced the spectral effect model of the spectral variation with regard to remote sensing classification.