不同分辨率遥感图像混合像元线性分解方法研究  

Research on LSMM of Remote Sensing Images Based on Different Resolution

作　　者：胡潭高[1] 张锦水[1] 贾斌[1] 潘耀忠[1] 董燕生[1] 李乐[1] 

机构地区：[1]北京师范大学地表过程与资源生态国家重点实验室,北京师范大学资源学院,北京100875

出　　处：《地理与地理信息科学》2008年第3期20-23,40,共5页Geography and Geo-Information Science

基　　金：国家高技术研究发展计划项目(2006AA120103、2006AA120101);教育部新世纪人才支持计划资助项目

摘　　要：混合像元的存在是传统像元级遥感分类和面积量测精度难以达到实用要求的主要原因,为了提高遥感应用精度,须解决混合像元分解问题。传统的方法主要通过改进分解模型提高分解精度,该文在不改变线性分解模型的条件下,分析不同分辨率尺度对于线性分解精度的影响。实验中运用像元合并的方法,得到不同分辨率的TM系列遥感图像,分别选取植被、裸地、水体3种典型地物进行线性分解;以分辨率更高的Quickbird图像分类结果作为真值进行精度评价。实验结果表明:随着图像分辨率的降低,植被的RMSE值不断缩小,在30 m分辨率尺度上均值为0.36,在150 m尺度上均值为0.17,分解精度提高了1倍左右;但随着分辨率进一步降低,由于混合像元现象加剧,RMSE值上升,分解精度随之降低。The accuracy of traditional classification method based on remote sensing image is difficult to achieve practical requirements because of the existence of mixed pixel. In order to improve the accuracy of remote sensing applications, it must solve the problem of spectral unmixing. Traditional methods improve the accuracy of decomposition through the improved spectral mixed model. In this paper,the research was based on LSMM and did not change the conditions of LSMM, and the method of pixels merger was used to get TM series of remote sensing images. The decomposition result was also compared with the classification result from Quickbird image. The experiments showed that with the reduction of image resolution, the RMSE of vegetation became smaller. In a resolution of 30 m, the average of RSME was 0. 36, and in a resolution of 150 m, the average of RSME was 0. 17, the accuracy was improved nearly 1 times. However, as the resolution further reduced, the number of mixed pixels was increased,and accuracy began decreased. In addition,as the conditions changed, how to choose an appropriate resolution is also necessary to studied in the future.

关 键 词：线性光谱混合模型 像元合并 空间分辨率 终端像元 均方根误差 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]