Font Size: a A A

Research Of Vegetation Hybrid Inversion With Hyperspectral Data Based On Transform Alogrithm

Posted on:2014-05-27Degree:MasterType:Thesis
Country:ChinaCandidate:M WangFull Text:PDF
GTID:2180330422474545Subject:Geodesy and Survey Engineering
Abstract/Summary:PDF Full Text Request
Leaf area index (LAI) is one of the important vegetation structural parameters,hyperspectral remote sensing makes it possible to estimate LAI in large area. However,the data redundancy caused by large number of bands and strong correlation amongthem neither wastes inversion time nor reduces data use efficiency. This paper appliedthe comprehensive inversion algorithm, in this method the PROSAIL model wascombined with neural network. Considering the transfer domain, principle componenttransform and partial least square transform were use here to reduce hypersprctralremote sensing data dimensionality, and then the transformed feature information wasused for retrieving LAI, further, compared to the full band inversion and feature bandinversion, and also analyzed the factors influencing the hyperspectral remote sensinginversion. The specific contents and conclusions are as follows:(1) The spatial dimensionality of training data influences the inversion results. Thisstudy proposed to transform original data with principal component analysis(PCA)and partial least square(PLS)regression, and applied the feature information forinversion,compared to the full band inversion and feature band inversion, theresults are as follow:PCA and PLS can reduce data dimensionality, inversion precision withtransformed data shows to be better. The data transform method reduces thecorrelation of inversion error equation and increases the convergence andefficiency.When the soil information is unknown, inversion based on transformed domaincan avoid the LAI negative phenomenon, which appears with feature banddata.PLS extracts the primary information which can replace the independentvariable and dependent variable, taking into account the correlation betweenthe independent variable and dependent variable. Transform the full band databy PLS algorithm, then construct the inverse compute model with neuralnetwork to retrieve LAI, it shows that the inversion precision is higher than thelinear regression model based the partial least squares regression analysis.(2) Noise makes the simulation closer to the true value of the spectrum. In inversion based on PCA transform, LAI was inverted by the inversion model constructed withthe noise simulation data. When the soil information is unknown, inversionprecision with noise simulation data is higher than noise-free data, and reaching theoptimal accuracy needs less principle component; when the soil information isknown, inversion with noise data reaches optimal accuracy with less principlecomponent too, but the optimal retrieve accuracy is without increase.(3) PCA and PLS have the anti-noise capability. Researches shows that inversionaccuracy with noise simulation data transformed by PCA and PLS is higher thanboth full band noise and noise-free data. Especially after PLS transform the optimalinversion results are significantly better than full band.
Keywords/Search Tags:hyperspectral remote sensing inversion, leaf area index, PROSAIL model, neural network, principle component analysis, partial least squaresregression analysis
PDF Full Text Request
Related items