Hyper-spectral Prediction Model Of Soil Moisture Content Based On Fuzzy Recognition

As the material basis for plant survival,soil moisture plays an important role in the distribution of plant species on land surface.Therefore,the prediction of soil moisture content is crucial for the growth and development of crops.From the traditional laboratory testing methods to the current use of hyper-spectral remote sensing technology,the quantitative prediction of soil moisture has undergone great changes.Hyper-spectral remote sensing technology has many unique advantages,including multiple band,high resolution and large amount of data,which make the research work in real time,fast,efficient and non-destructive,and provide an effective way for the inversion of soil trait indicators.The soil moisture content and spectral reflectance data are influenced by many factors,leading to the uncertainty of the data in the size division,which is the inevitable randomness and fuzziness.The fuzzy theory is mainly used to describe the uncertainty of research objects.Therefore,the use of fuzzy theory for soil moisture prediction has a certain theoretical basis.In this study,94 brown soil samples from Tai'an City,Shandong Province were taken as the research object.The outdoor reflectance spectral data and laboratory test data of moisture content were acquired for the collected samples.First,the spectral reflectance data were preprocessed,including breakpoint correction,smooth optimization,and abnormal sample rejection.On this basis,a variety of mathematical transformations were performed on the spectral data and the spectral reflectance characteristics of soil moisture were analyzed.Understand the sensitive bands of water content in all transform spectra,and the characteristic factors were determined based on single correlation analysis.Finally,the fuzzy recognition prediction model for soil moisture content based on hyper-spectral was established,including the variable fuzzy set prediction model and the semi-supervised fuzzy recognition prediction model,and other common prediction models were established for comparative analysis.The main research contents and results are as follows:(1)Extraction of spectral characteristic factors of soil moisture content.Through the transformation processing of spectral reflectance data,the correlation between each transform spectra and soil moisture content was determined by the single correlation analysis method,and the characteristic factors were selected according to the principle of maximum correlation principle.The two sets of data were selected as the characteristic factors,including 669 nm,762nm,1022 nm,1235nm,2060 nm under the first-order differential of square roottransformation,and 655 nm,1235nm,1497 nm,1677nm,2059 nm under the first-order differential of logarithmic reciprocal.The correlation between these spectra and water content is good.(2)Establishment of hyper-spectral fuzzy recognition model for soil moisture content.Based on the feature factors,the fuzzy recognition model and other common models were established for moisture content prediction.The results show that the variable fuzzy set model has the best prediction effect based on the selection of the characteristic factor in the firstorder differential of square root transformation.The average relative error and the coefficient of determination were 2.761% and 0.972,respectively.The posterior difference ratio and the small probability error were 0.284 and 1,respectively.Among other commonly used prediction models,the support vector machine model has relatively good prediction effect,with accuracy index values of 9.642%,0.957,0.208 and 1,respectively.It was slightly worse than the variable fuzzy set model.The semi-supervised fuzzy identification model has the best prediction effect based on the first-order differential of logarithmic reciprocal transformation.The average relative error and the coefficient of determination were 4.406% and 0.977,respectively.The posterior difference ratio and the small probability error were 0.384 and0.944,respectively.The accuracy of the decision tree model was better than that of other common prediction models,and the precision index values were 4.940%,0.925,0.326 and0.944 respectively.Considering the four evaluation indexes,the prediction effect of other models is relatively poor.It shows that the use of fuzzy recognition model to predict hyperspectral soil moisture based on certain mathematical transformation spectral data has certain feasibility and effectiveness.