基于线性光谱混合模型的城市下垫面分类影响因素分析

基于Hyperion高光谱与ALOS全色波段的融合图像,运用线性波谱分离模型,对影响城市下垫面分类的因子进行定量分析。结果显示:随着空间分辨率的递增,融合图像光谱保真性呈下降之势,并呈现出阈值现象,但所有相关系数值都大于0.90,表明以融合图像进行线性波谱分离提取城市下垫面具有可靠性;下垫面分类总精度则整体明显下降,Kappa系数值也几乎单调减少。大气校正后下垫面分类的总精度和Kappa系数都明显高于大气校正前;当丰度分割值由10%增至60%时,生成的下垫面分类图像的精度单调降低,表明应慎用50%阈值分割法。

Quantitative Study on Factors for Urban Underlying Surface Classification Using Linear Spectral Mixing Model

Based on fusion images of Hyperion and ALOS,using linear spectral mixing model,quantitative analysis on fusion effect and its factors was explored,which included atmospheric correction via FLAASH in ENVI,spatial resolution of image,threshold value for classification of rule images.The results showed that much information would be lost with spatial resolutions becoming coarser,as well as a threshold value existed.Still,correlative coefficients between raw images and coarser resolution images were all larger than 0.90 which indicated images kept their raw spectrum and could be used to identify surface of land.Meanwhile,total accuracy and Kappa coefficient presented downward trend.Total accuracy and Kappa values for classification corrected by FLAASH were larger than that for uncorrected images,while classification map was turned into pieces which were not in accord with the actual condition.Segment threshold of fraction was one of key factors when fraction images were separated into patches to create classification images.While threshold value was from 10% to 60%,total accuracy of classification maps displayed the opposite trend to decrease.The spectrum in a pixel within the area ratio may tend to be balanced,and there was not absolutely dominant spectrum in the pixels.This revealed the value of 50%,commonly adopted,should be used on condition.And Here the different thresholds should be given for different objects.