Research On Soil Moisture Content Estimation Algorithm Based On Hyperspectral Unmixing

Soil Moisture Content(SMC)plays an important role in a variety of scientific fields,including agriculture,hydrology,meteorology and ecology.In the past two decades,with the progress of earth observation technology,it has been shown that soil moisture content can be measured by electromagnetic wave to some extent,and remote sensors cannot directly measure soil moisture content.Therefore,it is of great significance to establish an inversion model of soil moisture content.The hyperspectral soil moisture content inversion technique mainly relies on reflectivity method or vegetation index method.Reflectivity method refers to the reflectivity of "bare soil",while vegetation index method refers to the reflection law of "vegetation" in different wave bands.However,for hyperspectral images,especially for satellite remote sensing hyperspectral images,it is difficult to obtain the reflectivity of pure vegetation or bare soil due to the ubiquitous mixed pixels limited by spatial resolution.To solve this problem,this thesis first decomposes the hyperspectral images and then calculates the vegetation index to build an accurate soil moisture content inversion model.The main contents are as follows:Firstly,in view of the current non-negative matrix-based unmixing algorithm is sensitive to the initial value and unstable.This thesis proposes an improved spatial spectral preprocessing algorithm to select relatively pure pixel and optimize the initial endmembers and the abundance value of the hyperpsectral unmixing algorithm.Then,according to the spatial relation of abundance sparsity and abundance,the low-rank constraint and the Laplace regular term constraint based on graph are introduced.Secondly,this thesis use the accurate vegetation reflectance after spectral unmixing to obtain vegetation indices,the grey correlation method was used to analysis of the correlation degree between vegetation indices and soil moisture content,the partial least square method was used to select the vegetation indices,and established a linear regression model of soil moisture content on vegetation indices and good results have been achieved.