Studies On Soil Moisture Hyper-spectrum Characteristics And Estimating Model

Soil moisture is one of the important components in crop growing, rapid and accuratedetermining farmland soil moisture has important meaning for developing precisionagriculture and improving field’s scientific management. Traditional soil moisture measuringmethods, such as gravimetric, neutron, tension-meter, is thinned in point, slowed and limitedin scope. Conventional optical remote sensing is lowed in band and spectral resolution,having much error for large area of soil moisture monitoring. At present, Hyper-spectralremote sensing which has higher spectral resolution and multi-band have become one of theimportant ways to measure soil moisture, one important link is estimation model. This articlebased on the Hyper-spectral data band range from350nm to2500nm of the84soil samples inShanxi province Hengshan country, discussed the soil moisture estimation model ofhyper-spectral with multiple linear regression, BP neural network, and fuzzy recognition.Firstly, this article analyzes the hyper-spectral characteristics of different soil types anddifferent soil moisture. Different soil type with different reflectance, along with Conton sand,Loessial soil, Aeolian sandy soil and sand soil reflectance in turn decrease. Soil moisture shortfor10%, soil moisture hyper-spectral characteristic is more sensitive in band1350~1400nm,while more than10%,1880~1920nm band is more sensitive. Secondly, soil reflectance dataare transformed by a number of first-order differential, envelope and anti-envelope areremoved to get the extraction of features, according to related principle extraction to chooseinversion factor. Reflectivity logarithm first order differential transformation in1432nm,1546nm,1760nm,1916nm,2060nm bands are selected for inversion factors.1414nm and1920nm position of envelope and anti-envelope remove transformation the largest position,depth area/difference area, high area/area, depth area/height area selected for inversion factors.Then, the soil moisture estimation hyper-spectral models are discussed by multiple linearregression, BP neural network, and fuzzy recognition. Inversion factors of Envelope andanti-envelope extract of position1920nm has the highest accurate, fuzzy recognition model that decided coefficient is higher than0.98and average relative error less than10%is betterthan other models. Finally, Hyper-spectral data obfuscation system is developed, which canprovide technical support for mass hyper-spectral data automation processing and fuzzyrecognition.