基于遥感影像的大区域植被类型样本快速提取方法研究

目的] 利用遥感影像的宏观性，基于植被分类资料数据，依据实验区域遥感影像及衍生影像本身特点，实现大区域样本快速提取。方法] 实验参考1：100万植被图、WESTDC中国土地覆盖图，结合实验区域2001年MODIS时序NDVI影像的非监督分类结果，利用矢、栅数据的空间特征，将实验影像非监督分类的类型信息关联为随机样点属性，依据该属性中包含的非监督分类类型数和各类型的样点比例，对比类别间样本可分离性指标、标准差变化，实现样本纯化。结果] 纯化后的植被样本与WESTDC中植被空间分布基本一致，主要植被类型空间分布精确程度为84.82%。将纯化前后的样本输入最大似然分类器，总体分类精度提高了32.52%。结论] 该采样方法适用于宏观大区域植被样本数据的快速提取。同时，节省了大区域植被类型调查消耗的人力物力资源和时间，提高了采样效率。

Quick Sampling Method for Large Area Vegetable Types Based on Remote Sensing Images:A Case Study for Cold Temperate Coniferous Forest Region

Objective] lize the quick access of the separable and representative training samples on large area remote sensing images.Method] Based on the macroscopical character of remote sensing images, referring to the 1:1 000 000 vegetation map and China land cover map (WESTDC), and combining the unsupervised classification results of experimental region from the 2001 MODIS time series NDVI images, the spatial features of vector and raster data were used and associated the unsupervised classification results as the random sample properties. Based on the unsupervised classification types and the random point ratio of different classification types, the sample purification was realized. By comparing the difference of the standard deviation and the separability index before and after the sample purification of different vegetable samples, the points which had relatively larger contribution to the supervised classification were saved.Result] The results showed that the land cover spatial distribution in survey area was consistent with the WESTDC map, the overall accuracy of main types was 84.82%. By using the purred samples and applying the maximum likelihood classification method, the overall accuracy was 32.52% higher than that from initial samples.Conclusion] The sampling method is suitable for quick access of remote sensing images of large area and could reduce the time, manpower and material costs.