考虑边界区域的地表覆盖分类精度评价方法

发布时间：2018-01-29 18:10

本文关键词： 地表覆盖分类精度评估分层抽样边界区域内部区域　出处：《中国科学:地球科学》2016年11期 　论文类型：期刊论文

【摘要】：地表覆盖作为地表状况的直接反映,与全球物质能量循环、气候变化以及人类社会经济活动等密切相关.地表覆盖分类数据通常利用遥感图像进行分类得到,其数据质量对全球环境监测及其变化研究与决策支持等过程有重要的影响.因此,对分类结果进行精度评估是地表覆盖分类的重要步骤.由于影像上不同地物类型边界处往往存在较多的混合像元,使得分类时处于边界处的像元分类精度相对低于内部像元,直接导致了类内自身的分类精度异质性增大,分类精度评估的不稳定性也随之增大.本文基于传统的分层抽样(stratified sampling,SS),提出了一种考虑边界区域的分层抽样方法(stratified sampling considering edges,SSCE).理论推导及实验结果表明,SSCE对总体精度(overall accuracy,OA)和kappa估计的准确性和稳定性均优于SS方法,主要结论为:在样本点总数较少的情况下,SSCE对分类精度估计的准确性高于SS方法;在相同的容许误差下,SSCE方法所需的样本数小于SS方法;分类图像中类内边界区域与内部区域分类精度的差异越大、其面积比例越均衡,SSCE方法越能准确地进行精度估计.因此,本文提出的SSCE方法能够以更少的样本获取代价、更高的准确性和稳定性对地表覆盖数据进行分类精度的评估.
[Abstract]:Surface cover, as a direct reflection of the surface condition, is closely related to the global material and energy cycle, climate change and human social and economic activities. Surface cover classification data are usually obtained by remote sensing images. Its data quality has an important impact on the process of global environmental monitoring and its change research and decision support. The accuracy evaluation of classification results is an important step in land cover classification, because there are many mixed pixels at the boundary of different types of objects in the image. The classification accuracy of the pixel at the boundary is lower than that of the internal pixel, which directly results in the heterogeneity of the classification accuracy within the class. The instability of classification accuracy evaluation also increases. This paper is based on the traditional stratified sampling SSSS. A stratified sampling considering edges method considering boundary region is proposed. The theoretical derivation and experimental results show that SSCE has overall accuracy accuracy. The accuracy and stability of kappa estimation are better than SS method. The main conclusions are as follows: the accuracy of classification accuracy estimation is higher than SS method when the total number of sample points is small; The number of samples required for the SSCE method is smaller than that for the SS method under the same allowable error. The greater the difference between the classification accuracy of the inner boundary region and the inner region in the classification image, the more balanced the area ratio is, the more accurate the accuracy estimation can be by using the SSCE method. The SSCE method proposed in this paper can evaluate the classification accuracy of land cover data with less cost and higher accuracy and stability.
【作者单位】：北京师范大学地表过程与资源生态国家重点实验室;北京师范大学全球变化与地球系统科学研究院;
【基金】：国家高技术研究发展计划项目(编号:2013AA122802)资助
【分类号】：P237
【正文快照】： 国家高技术研究发展计划项目(编号:2013AA122802)资助1引言地表覆盖是自然过程与人类社会经济活动过程相互作用的产物,其变化直接影响到地表水分、能量和碳循环.全球地表覆盖的空间分布与动态变化制图已成为全球变化研究、地理国(世)情监测、可持续发展规划、土地资源管理等领

【相似文献】