摘要
针对线性混合像元分解(Linear Spectral Unmixing,LSU)在端元(Endmember)个数不变情况下常会出现端元分解过剩现象导致分解结果精度不高的问题,以地物分布的聚集性特征为基础,提出了基于格网的变端元线性混合像元分解(Dynamic Endmember LSU,DELSU)方法。以冬小麦为研究目标,采用Landsat TM图像为实验数据、高分QuickBird图像目视解译冬小麦结果为真值精度评价数据,利用本文提出的DELSU方法进行冬小麦提取。实验结果表明:该方法比最大似然方法、LSU方法更能准确地获取冬小麦面积,在一定程度上吸收了传统分类方法的优点,提高了目标地物的测量精度;同时作为一种改进的LSU方法也适用于其他土地利用/覆盖类型的测量。
Linear spectral unmixing （LSU）is the most common method for solving mixed pixel problem; nevertheless, if the number of the pixels＇ endmember is regarded as unchangeable, the traditional pixel unmixing algorithm cannot attain a good result. In the light of the characteristic that pixels usually have the same composition as their neighboring pixels, the authors proposed a grid - based dynamic endmember linear spectral unmixing （DELSU） model. The land cover classification experiment was conducted with the adoption of the Landsat TM image as the experimental data. The abundance map of winter wheat derived from the visual interpretation of the QuickBird image was used for accuracy evaluation. The experimental results show that the use of the DELSU model could extract the area of winter wheat at a precision higher than that of the traditional maximum likelihood method and the LSU model. This model absorbs the traditional classification advantages and improves the measurement accuracy of the target features. As an improved method of LSU, DELSU is also applicable to the measurement of other land use/cover types.
出处
《国土资源遥感》
CSCD
2011年第1期66-72,共7页
Remote Sensing for Land & Resources
基金
农业部资源遥感与数字农业重点实验室开放基金项目(编号:RDA0807)
国家高技术研究发展计划资助项目(编号:2006AA120103、2006AA120101)共同资助
