Hyper-spectral Estimation Model For Soil Organic Matter Content Using Grey Scale Of Grey Number

Soil organic matter is an important component of soil and an important indicator for evaluating soil fertility.Rapid and efficient acquisition of organic matter information is of great practical significance for quantitative fertilisation and the development of precision agriculture.At present,the use of hyper-spectral remote sensing technology to estimate soil composition has become a hot spot.However,due to the diversity and complexity of soil spectral influencing factors,the accuracy of hyper-spectral estimation of soil organic matter content needs to be improved.Therefore,in this paper,76 brown soil samples collected from Zhangqiu District of Jinan City were used as experimental data,on which the hyper-spectral grey number grey estimation model of soil organic matter content was established by combining grey system theory and statistical analysis.The main contents and conclusions of the study are as follows.（1）The soil spectral characteristics of brown soils in Zhangqiu District were revealed.The nine-point weighted moving average method was first used to smooth out the noise in the soil spectral curves,then the samples were divided into four groups according to the increasing order of organic matter content,and the average value of each group of samples was taken to obtain the soil spectral curves with different organic matter contents and analyse their spectral characteristics.The results showed that the spectral curves showed an increasing trend in the 350nm-1450nm band,with the fastest increase in the 550nm-850nm band and a slightly slower increase in the rest of the band;in the 1450nm-1800nm band,the spectral curves first increased and then decreased,with small changes and stable curves;in the1900nm-2100nm range,the spectral curves showed an obvious increasing trend In the2200nm-2500nm range,the spectral curve shows a clear decreasing trend;the outdoor spectrum is noisier at the three water absorption valleys of 1400nm,1900nm and 2200nm.With the gradual increase of soil organic matter content,the soil spectral reflectance showed a decreasing trend.（2）Soil organic matter spectral characteristic factors were extracted.Nine transformation methods such as logarithmic and inverse were used to process the soil reflectance,calculate the correlation coefficient between the spectral transformation value and the soil organic matter,and select the characteristic factors according to the principle of great correlation.The results showed that the log-inverse first-order differential transform was more effective.Therefore,the first-order differential transform values of the logarithmic inverse at 530nm,562nm,1507nm,2035nm and 2108nm bands were selected as the soil organic matter spectral feature factors,and their correlation coefficients were 0.68,0.70,0.68,0.77 and 0.68respectively.（3）Two hyper-spectral grey number grey scale estimation models of soil organic matter content were developed.Firstly,the estimators or organic matter contents of known models are ranked from smallest to largest,and then the sliding variances of the corresponding organic matter contents or estimators near the points are calculated in turn.Secondly,the sliding variance is considered as a direct representation of the greyness of the grey number and is converted into the relative greyness value of the grey number.In the third step,a correction model for the estimation factor is constructed based on the greyness,and then a grey number greyness estimation model for organic matter content is built and compared with commonly used modelling methods（BP Neural Network,Random Forest,Multiple Linear Regression and Support Vector Machine,etc.）.The results showed that the two models based on grey number greyscale and three-dimensional greyscale could effectively improve the estimation accuracy,in which the determination coefficients R~2of the 12 tested samples were0.967 and 0.929 respectively,and the mean relative errors were 7.298%and 6.830%respectively.In contrast,the coefficients of determination R~2for Multiple Linear Regression,BP Neural Network,Support Vector Machine and Random Forest were 0.680,0.812,0.684and 0.761 respectively,with mean relative errors of 13.307%,14.084%,10.400%and10.912%respectively.The study shows that the hyper-spectral grey number grey scale estimation model for soil organic matter content is feasible and effective.