摘要
土壤含水量的高光谱反演是当今研究的热点。以土壤多样化的陕西省横山县为研究区,通过野外采集土壤样品,室内利用ASD Field Spec FR地物光谱仪测定土壤样品光谱,采用称重法计算出土壤样品含水量,并分析了不同含水量土壤样品的光谱特性。针对土壤含水量光谱反演中光谱反演因子的构建问题,在研究一阶微分(FD)-主成分分析(PCA)、小波包变换(WPT)-FD-PCA反演输入因子生成方法及存在的不足的基础上,提出了基于谐波分析(HA)的WPT-FD-HA-PCA的反演输入因子构建方法。以上述三种反演输入因子为基础,建立了土壤含水量反演的FD-PCA-反向传播(BP)、WPT-FD-PCA-BP、WPT-FD-HA-PCA-BP三种BP反演模型。通过比较土壤含水量实测值与三种反演输入因子的反演结果,得出WPT-FD-HA-PCA-BP模型的反演精度最高,决定性系数R2达到0.9599,均方根误差为1.667%,其反演结果明显优于其他两种模型。这表明通过WPT和谐波分析能有效地抑制光谱噪声并压缩信号,在一定程度上明显提高了土壤含水量反演精度。
Hyperspectral inversion of soil water content is a current hot research topic.Hengshan County of Shaanxi Provice,which has diverse soil types,is taken as study area.Soil samples are collected in the field and their spectra are tested indoor using ASD Field Spec FR ground object spectrometer.Moreover,the soil water content is calculated by weighing method and the spectral features of the soil samples with different water contents are analyzed.For the construction issue of factors in the spectral inversion of soil water content,and based on the study of inversion input factor generation method and the existing problems of first order differential（FD）-principal component analysis（PCA）,wavelet packet transform（WPT）-FD-PCA,the method of constructing the inversion input factor of WPT-FD-HA-PCA based on harmonic analysis（HA）is proposed.On the basis of the three abovementioned inversion input factors,three back propagation（BP）models of soil water content（FD-PCA-BP,WPTFD-PCA-BP,WPT-FD-HA-PCA-BP）are constructed.Comparison between the measured values of soil water content and the inversion values of the three BP models shows that the inversion accuracy of WPT-FD-HA-PCA-BP model is the highest.The coefficient of determination（R2）and root mean square error between measured value and inversion value is 0.9599 and 1.667% respectively,and it performs better than the other two models.The results show that WPT and HA can effectively suppress the spectral noise and compress the signal,and to some extent,the inversion precision of soil water content is improved obviously.
作者
姜雪芹
叶勤
林怡
李西灿
Jiang Xueqin;Ye Oin;Lin Yi;Li Xican(College of Surveying and Geo-Informatics, Tongji University, Shanghai 200092, China;College of Information Science and Engineering, Shandong Agricultural University, Taian, Shandong 271018, China)
出处
《光学学报》
EI
CAS
CSCD
北大核心
2017年第10期292-302,共11页
Acta Optica Sinica
基金
上海市科委项目(13231203602)
关键词
遥感
土壤含水量
小波包变换
谐波分析
主成分分析
反向传播神经网络
remote sensing
soil water content
wavelet packet translation
harmonic analysis
principalcomponent analysis
back propagation neural network
