Research On Extraction Of Water Body And Sediment Concentration Distribution Inversion Based On HSI Hyperspectral Remote Sensing Data

In the investigation of water resources, water resources macro functionmonitoring and assessment of natural disasters, using remote sensing image to extractwater body information has been widely used. Water body information extraction arethe traditional remote sensing image based on Landsat TM, SPOT-5and MODISdata, HSI (the Hyper Spectral Imager) data is the first domestic space hyperspectraldata, domestic research on HSI data in the area of water body extraction is relativelyless. To the silver sea area in North Sea Area in Guangxi and Hepu county in2014onJanuary15, one scene HSI data has been for the study area, mainly searching on thefollowing work:(1) HSI data introduction and pretreatment research: This paper introduces thefeatures of the HSI data, and the HSI data including data format conversion, radiationcorrection (FLAASH atmospheric correction), the preprocessing work of the poorquality of band to remove fine correction and collections.(2) A spectral characteristics analysis and band selection: after datapreprocessing, get the HSI data including function of water body, the typical featuresof spatial information, analysis of spectral curve; through the analysis of the spectruminformation, using adaptive band selection method;through the analysis of thedifferent feature class, based on the spectral band selection of separability distancemethod; Comprehensive two methods, select the best bands.(3) The method comparison of the study area of water body informationextraction: Respectively using supervised classification method (maximum likelihoodand support vector machine, unsupervised classification (ISODATA and K-Means),Index method (NDWI) and the decision tree method (CART), to extract the water ofthe study area, through the comparative analysis, it shows that accuracy of thedecision tree algorithm in the study to the highest value.(4) The purpose of study on reflection spectral characteristics of suspendedsediment in water is toanalyze numerical relationship betweenspectral reflectance ofwater and suspended sediment concentration by using a linear model, exponentialmodel, and logarithmic model. Exponential model that is compared countis moresuitable the study area. Finally, the paper summarize the work and discusses the research directionafterwards.