Study On Estimation Of Soil Manganese Content Based On Experiment Spectra

With the modernization of agricultural production and the neglect of soil environmental capacity, the soil heavy metal pollution is more serious. Depends on its high spectral resolution, hyperspectral remote sensing has the potential of quantitative acquisition of soil chemical component. Using reflectance spectroscopy to estimate the soil heavy metal element content has far-reaching significance, it will provide preliminary theoretical research in the utility of soil remote sensing. The research uses Jiangsu soil samples, studies the relationship between different soil particle size (0.25mm,1mm,2mm) and their reflectance spectra, obtains the suitable particle size of laboratory spectral measurements. Then using principal component regression, partial least squares regression, support vector machine regression, extreme learning machine algorithm to establish the manganese content in soil reflectance, the first derivative, reciprocal logarithmic estimation model. Use of AVHRR and MERSI sensor simulation spectra, further explore the feasibility of using multi spectral sensor simulation reflectance to estimate manganese content. The main conclusions are as follows:The results show that, there is a negative correlation between soil particle size and the spectral reflectance. Establish the model of relationship between the reflectivity of different particle size and manganese content by using principal component regression. The particle size which is2mm in the estimation of manganese content obtains a relatively high precision. Different transformations of spectra have a greate impacte on the estimation results. For different estimation models, the first derivative transformation is better than the reflectance and reciprocal logarithmic transformation. The model reaches the extremely significant level. Compared with the methods of principal component regression and principal component regression, using the methods of the support vector machine regression and extreme learning machine algorithm to estimate the manganese content in soil achieve relatively high accuracy. Therefore, the first derivative was a kind of effective spectral transform method. The methods of support vector machine regression and extreme learning machine algorithm were effective methods to increase the accuracy and the stability of the model. Used the first order derivative to transform, then established the model by the methods of support vector machine regression and extreme learning machine algorithm. The correlation coefficient of test samples is0.731and0.653respectively. The root mean square error is0.042g/kg and0.046g/kg respectively. Ratio of Prediction to Deviation is1.359and1.252respectively. Transforming the hyperspectral data into multi spectral data based on spectral response function to predict the manganese content in soil is feasible. The method of the support vector machine regression and extreme learning machine algorithm get higher forecasting accuracy. The accuracy of the regression model which uses15bands from the data of MERSI is generally higher than uses3bands from the data of AVHRR.