从TM遥感影像提取针叶林、阔叶林和低矮植被的丰度信息,对森林资源调查和生态建设具有重要意义。为改善线性混合像元分解的精度,除了经过FLAASH大气校正、MNF变换、PPI计算和人-机交互可视化端元选择等一般步骤外,尝试增加了地形校正并修复辐射极值、NDVI阈值法掩膜处理和反射率归一化处理3个环节。与此同时,进行了相似线性分解的简单实验,结果表明：增加了这3个处理环节可以改善线性分解的精度,RMS分量图显示均方根误差图的平均值由0.009 46降到0.004 91,均方根误差小于0.007 75的像元数占误差图全部像元数的百分比由55.61%上升到81.36%。选择野外随机均匀分布的60个样本点对模拟结果进行了线性拟合检验。实验二阔叶林、针叶林和低矮植被丰度模拟值和实测值之间的相关系数R2比实验一分别提高了0.211 6、0.045 6和0.130 7。%Abstraction of the abundance information of broadleaf forest,coniferous forest and srubby vegetable from TM remote sensing images has great importance to forest resource survey and ecology construction.To improve the accuracy of extraction of the vegetation abundance information with linear mixed model,in this theses,not only some general methods that include Minimum Noise Fraction（MNF）,Pixel Purity Index（PPI） and n-D Visualized had been adopted,but also three other data processing links that include topographic correction（including replacing the extreme value of the images after topographic correction）,masking some components using suitable NDVI threshold value and reflectivity normalization had been added.Meanwhile,the simple experiment,which was similar to the previous linear decomposition experiment.The results showed that three other data processing links added can improve the accuracy of extraction of the vegetation abundance information,and also showed that the mean value of RMS（root mean square error） was decreased from 0.009 46 to 0.004 91 and the percentage of the numbers of pixels which values were less than 0.007 75 in the whole numbers of pixels in accordance with the RMS abundance image was increased from 55.61% to 81.36%.The 60 sample points which were randomly sampled and uniformly distributed in the field survive were used in linear fitting test of the simulation results.The correlation coefficient R2 between the simulation value and the measured value of broad-leaved forest,coniferous forest and low vegetation in Experiment 1 is up by 0.211 6、0.045 6 and 0.130 7 than that in Experiment 2.
展开▼
