Hyper-Spectral Estimation Model Of Soil Organic Matter Using Grey Information

Soil organic matter is an important part of soil information.Soil organic matter plays an extremely important role in soil fertility evaluation,sustainable development of agriculture and forestry,precision agriculture and other aspects.Hyper-spectral remote sensing,due to its advantages of multi-band and rich information,can provide technical support for rapid and large-scale information acquisition of soil organic matter.Since the complexity of the influencing factors,the accuracy of existing estimation methods is not ideal and is not fully utilize insufficient information.There for,this paper takes Zhang-qiu District,Jinan City,Shandong Province as the research area based on the data of 76 soil samples,which discusses the hyper-spectral information estimation model of soil organic matter from the perspective of information.The main contents and conclusions of the study are as follows.1.The soil spectral characteristics of the cinnamon soil in Zhang-qiu District were analysed.Firstly,the soil spectral curve was smoothed and denoised,and 76 soil samples were used to draw the grouping spectral characteristic curve,and then the spectral characteristics were analyzed.The results show that the soil spectral reflectance gradually decreased with the increase of soil organic matter content.The spectral curves of different organic matter contents of soil samples have the same change trend in the whole band range.In the range of 350 nm-1300 nm,the spectral curve mainly shows an up-ward trend,among which the 550 nm-850 nm band has the fastest rising speed,and the rising speed of other bands is slightly slower.In the range of 1450 nm-1800 nm,the spectral curve first rises and then falls.The change of the curve is small and the growth trend is stable.In the range of 1900 nm-2100 nm,the spectral curve shows a clear upward trend.In the range of 2200 nm-2500 nm,the spectral curve shows a clear down-ward trend.On the 1400 nm,1900 nm and 2200 nm bands,the spectral curve appears obvious trough period.2.The spectral characteristic factors of soil organic matter were extracted.Firstly,logarithmic,square root,differential and other methods are used for spectral transformation.Then,the correlation coefficient between soil organic matter and the transformed spectral reflectance is calculated,and the characteristic factors are selected according to the principle of great correlation and as discrete as possible between bands.The results showed that the first-order differential transformation effect was the best,which could effectively improve the correlation between spectral data and soil organic matter.Finally,the spectral transformation values of 560 nm,1470 nm,1658 nm,2059 nm,2122 nm,2229 nm and2321 nm bands were selected as spectral estimation factors,and the correlation coefficients were-0.6401,-0.6791,-0.7218,-0.7035,-0.6713,-0.6399 and 0.7561.3.The hyper-spectral estimation model of soil organic matter based on grey information was established.Based on the grey system theory and the method of increasing interest rate,the spectral grey information matrix was established by using the characteristic factors of soil organic matter,and the main gradient grey information of the information matrix was extracted to establish the soil organic matter content estimation model based on grey information.The positive and negative correlation is used to identify the main gradient information of the sample to be estimated,and the estimation is based on the main gradient information of the sample to be estimated.Finally,the estimation results are compared with common models such as multiple linear regression,BP Neural Network,Support Vector Machine and Decision Tree to verify the validity of this model.The results showed that the hyper-spectral grey information estimation model of soil organic matter content had high estimation accuracy.The mean relative error of modeling and testing samples were 5.2699% and 7.4837%,and the coefficients of determination were 0.9426 and 0.9143.The mean relative error of modeling samples of multiple linear regression,Support Vector Machine,BP Neural Network and Decision Tree were13.0600%,10.7272%,14.2881% and 8.7233%.The coefficients of determination were 0.6311,0.7027,0.6142 and 0.8514.The mean relative error of test samples were 9.4819%,10.7088%,12.5225% and 16.3770%,and the coefficients of determination were 0.7984,0.8376,0.8926 and0.5058.The research shows that the hyper-spectral estimation model of soil organic matter using grey information proposed in the paper is feasible and effective.