基于混合像元分解的土壤盐渍化遥感监测研究  

Research on Remote Sensing Monitoring of Soil Salinization Based on Mixed Pixel Decomposition

作　　者：刘娅 陈丹艳 LIU Ya;CHEN Dan-yan(Jinling Institute of Technology,Nanjing 211169,China;State Key Laboratory of Soil and Agricultural Sustainable Development,Nanjing Institute of Soil Science,Chinese Academy of Sciences,Nanjing 210008,China)

机构地区：[1]金陵科技学院软件工程学院,江苏南京211169 [2]中国科学院南京土壤研究所土壤与农业可持续发展国家重点实验室,江苏南京210008 [3]金陵科技学院园林园艺学院,江苏南京210038

出　　处：《金陵科技学院学报》2022年第3期15-23,共9页Journal of Jinling Institute of Technology

基　　金：国家自然科学基金青年项目(41601214);江苏省重点研发计划项目(BE2019378);金陵科技学院高层次人才科研启动基金(jit-b-202005,jit-b-201914);金陵科技学院科研孵化项目(jit-fhxm-201910)。

摘　　要：植被覆盖等外部因素导致的混合像元问题增加了土壤盐渍化遥感监测的难度,一直是土壤盐渍化遥感监测的难关。以Landsat TM5遥感数据为数据源,利用线性方程模型构建混合光谱矩阵,再通过非负矩阵分解的方法从混合光谱矩阵中分解出土壤和植被光谱,以减弱植被覆盖等外部因素对土壤光谱信息的影响。利用提取到的土壤光谱信息估算土壤电导率,以监测区域土壤盐渍化的程度。实验结果表明:对于校正前后的光谱,随机森林模型均取得较好的监测结果,100次随机抽样R_(p)^(2)值为0.60,RPD值为1.36。The mixed pixel problem caused by external factors such as vegetation cover increases the difficulty of remote sensing monitoring of soil salinization,which has always been the difficulty of remote sensing monitoring of soil salinization.Taking Landsat TM5 remote sensing data as the data source,the mixed spectral matrix is constructed by using linear equation model,and then the soil and vegetation spectra are decomposed from the mixed spectral matrix by non-negative matrix factorization to reduce the influence of external factors such as vegetation cover on soil spectral information.Using the extracted soil spectral information to estimate soil conductivity to monitor the degree of regional soil salinization.The experimental results show that the random forest model can obtain better monitoring results for the spectra before and after correction.The average R_(p)^(2) value of 100 random samples is 0.60 and the RPD value is 1.36.

关 键 词：土壤盐渍化 混合像元 线性混合模型 非负矩阵分解 土壤制图 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]