遥感图像分类精度的点、群样本检验与评估

Accuracy Assessment of Thematic Classification Based on Point and Cluster Sample

遥感专题分类结果在使用前,必须进行客观可靠的精度验证和分析,以保持遥感分类结果的可靠性。本文利用不同分辨率遥感数据获取的同一地区土地利用/覆盖信息,进行了简单随机抽样、系统抽样和分层抽样三种不同抽样组织方式下的点样本和群样本检验分析,评估了不同抽样方式下的点样本和群样本检验效果。研究结果表明:(1)抽样方式对遥感分类精度评价结果的影响是客观存在的,不同抽样方式下的点样本和群样本检验结果都存在一定的随机性,但同一种抽样方式下,点样本检验精度评估结果的波动幅度小于群样本检验,稳定性比群样本检验要好;(2)不同抽样方式下的多次点样本和群样本检验的平均精度检验结果基本上都能够反映分类图像的精度特征,其中,点样本检验中,分层随机抽样点样本检验效果较好;群样本检验中,系统抽样群样本检验和分层随机抽样群样本检验的效果优于简单随机抽样群样本检验。

In order to assure the application of thematic classification, it is very important and necessary to make a rigorous accuracy assessment. In this paper, we use sampling-methods of point and cluster sample to assess the accuracy of the same region＇ s land-use/land-cover thematic maps, which are derived from different resolution remote sensing data. Here, Sampling designs are consisted of simple random, systematic and stratified sampling. The results are as following. Firstly, the sampling design has great impact on the accuracy of remote sensing classification. There exists great randomicity on the result of points and cluster sample verification on the different Sampling design. On the same sampling design, the stability of point sample verification is higher than that of Cluster sample verification. Secondly, the average accuracy of different sampling designs of multi-point and multi-cluster sample verification can reflect the accuracy characteristic. During the course of point sample verification, stratified sampling＇ s error is lower than others＇. During the course of cluster sample verification, systematic sampling and stratified sampling＇ s accuracy are prior to simple random sampling＇ s.