Study On Hyperspectral Indirect Estimation Model Of Soil Moisture In Plough Layer

Soil moisture is an important part of the soil,an important indicator of land resource evaluation and crop growth monitoring,and has an irreplaceable effect on crop growth and development.Optical satellite remote sensing can only detect the water content of the soil surface layer,and cannot achieve rapid monitoring of the water content of the soil plough layer.While the traditional soil moisture measurement method,which is mainly based on the drying method,has high accuracy,but the measurement period is long,the process is complicated,it takes time and effort,and it is difficult to obtain data quickly and accurately.Hyperspectral remote sensing provides new technologies for quantitative monitoring of soil moisture content in large areas due to the advantages of high spectral resolution,multiple bands,and abundant information.Therefore,studying the spectral characteristics of the soil surface layer and establishing an indirect spectral estimation model of the soil water content of the soil plough layer are of great significance for the rapid monitoring of the soil water content of the soil plough layer by optical satellite remote sensing and the development of fine agriculture.This study takes Jiyang City,Shandong Province as the research area,and takes the water content of 85 soil samples and its indoor and outdoor reflectance spectra as research objects.Based on the correlation between the surface layer and the cultivated layer,different methods are used.Experiments were conducted to establish an indirect spectral estimation model of soil moisture content in the plough layer,and the effectiveness of the indirect spectral estimation model was proved by comparison between different models.The main research contents and conclusions are as follows:(1)Analysis of soil spectral characteristics and determination of sensitive bands and characteristic factors of soil water content in Jiyang CountyBy using a comparative analysis method,the surface spectrum of the soil and the spectrum of the cultivated layer are analyzed,and then the spectrum is transformed using 9 spectral transformation methods such as square,square root,logarithm,reciprocal,first-order differential,and combinations thereof.The correlation analysis between the spectrum and the soil moisture content determines the sensitive band of the soil moisture content,and the characteristic factors are extracted using the principle of maximum correlation.The results show that the outdoor spectrum of soil water content and cultivated layer water content in Jiyang County under different water content conditions have roughly the same spectral change trend.The spectral reflectance of the soil decreases with increasing water content,and the two are negatively correlated.Sex.After spectral transformation,the effects of square root and logarithmic transformation are better,and the maximum correlation coefficient is increased from 0.90 to0.94;while other transformation methods are not effective.(2)Two indirect spectral estimation models of soil water content in cultivated layers were established.By analyzing the relationship between the measured values of the surface layer and the cultivated layer soil,it is found that there is a significant correlation between the two.Based on the correlation between the topsoil and the topsoil,two indirect spectral estimation models of the topsoil soil moisture content were obtained.The first is to analyze the internal relationship between topsoil and topsoil moisture content to establish topsoil and topsoil The model of the relationship between water content,and then the estimated value of the surface soil water content spectrum is substituted into this relationship model to calculate the indirect estimate of the soil water content in the cultivated layer.Indirect correlation,establish an indirect estimation model of soil moisture in cultivated layer based on surface spectral characteristics.Six conventional estimation methods are used to bring in two indirect estimation models.The results show that the results of the two indirect estimation models are similar.Among them,the model using the multiple linear regression method has the best results.The determination coefficient R~2 of the first model is 0.7649,the average relative error is 19.52252%;the determination coefficient R~2 of the second model is 0.8235,and the average relative error is 17.6957%.Studies have shown that it is feasible and effective to indirectly estimate the soil moisture content of the cultivated layer using the surface soil spectrum.(3)An indirect spectral estimation model of cultivated soil water content based on a convolutional neural network was established.According to the second indirect estimation model idea,combined with the convolutional neural network model in deep learning,the indirect spectral estimation model of cultivated soil water content based on the convolutional neural network is established.The processed four-dimensional spectral information matrix is used as input.The estimated value of the soil water content in the plough layer is the output value,and the result is compared with the actual value of the soil water content in the plough layer.The results show that with the increase of the number of iterations,the estimation accuracy of the convolutional neural network is getting higher and higher.When the number of iterations N is 3500,the average relative error of the estimation result is3.1849%,and the determination coefficient R~2 reaches 0.9909.The research shows that it is feasible and effective to use convolutional neural network to indirectly estimate the soil moisture content in the plough layer,and it provides a new idea for the optical remote sensing technology to be used to monitor the soil moisture and nutrient content in the plough layer.