| Soil organic matter can promote the growth and development of crops,and is also often used in the assessment of soil fertility.Hyper-spectral remote sensing technology has many advantages such as high resolution,real time and fast,rich information,and so on,which has been widely used in soil organic matter monitoring and other fields,and plays an important role in the development of precision agriculture.However,due to the complexity of influencing factors,there are inevitably three kinds of uncertainties in soil spectral estimation,namely,randomness,fuzziness and greyness,leading to less than ideal estimation accuracy.Therefore,this paper takes Jiyang and Zhangqiu District of Jinan City as the experimental area,and establishes a hyper-spectral grey fuzzy estimation model of soil organic matter based on the collected data of 121 sandy and brown soil samples,and comprehensively applies statistical analysis,fuzzy identification theory and grey system theory.The main research contents and conclusions are as follows.(1)The soil spectral characteristics of sandy and brown soils were analyzed.Firstly,the soil spectral curves were smoothed and denoised,and after eliminating six abnormal samples,the remaining 115 soil samples were used to draw grouped spectral characteristic curves of sandy soil and brown soil,and then the spectral characteristics were analyzed.The results showed that the grouped spectral curves of the organic matter content of 47 sandy soils and 68 brown soils were basically the same;the soil spectral reflectance generally decreased gradually with the increasing organic matter content of sandy and brown soils;in the 350nm-1350nm band,the spectral reflectance curve increased continuously with the increase of wavelength;in the1500-1800nm band,the spectral curve generally showed a trend of first increasing and then decreasing and gradually stabilized;near the 1400nm and 1900nm bands,the spectral curve fluctuates drastically and there is a large noise interference due to the influence of moisture in the air;in the 1950nm-2100nm band,the spectral curve rises faster;in the 2200nm-2500nm band,the spectral curve decreases obviously.(2)The spectral estimation factors of soil organic matter were extracted and corrected.Firstly,logarithmic,inverse,square root and other methods were used to transform the spectra,and the characteristic factors were selected based on the principle of great correlation and as discrete as possible between bands,and then the generalized grey scale of interval grey number was used to construct the correction model of estimation factor.The results showed that the first-order differentiation of logarithmic inverse and the first-order differentiation of square root inverse transformations were more effective,and the correlation coefficients between the corrected estimators and soil organic matter were significantly improved.Therefore,the transformed values of the logarithmic inverse first-order differential transform at 524 nm,569nm,1638 nm,2063 nm,2107 nm,2126 nm bands and the transformed values of the square root inverse first-order differential transform at 854 nm were selected as the spectral characteristic factors of soil organic matter,and their correlation coefficients were 0.7407,0.7006,0.6955,0.7782,0.7654,0.7687 and 0.7050,respectively.After the correction of the estimation factors,the correlation coefficients of the seven spectral estimation factors of the modeled samples and the soil organic matter were 0.8102,0.7888,0.7459,0.7175,0.8131,0.8159 and 0.8120,respectively.(3)A hyper-spectral grey fuzzy estimation model of soil organic matter was established.Firstly,the correction model of estimation factors was used to correct the estimation factors of the modeled samples,and the generalized greyness of the estimation factors of the test samples was calculated using the forward and reverse grey correlation and the estimation factors were corrected.Secondly,a grey fuzzy estimation model of soil organic matter is established based on the fuzzy integrated prediction model,and the estimation model is optimized by adjusting the fuzzy classification number;another grey fuzzy estimation model of soil organic matter with self-feedback is constructed based on the self-feedback fuzzy identification method,and the model is optimized by cyclic iteration of weights.Finally,the estimation results of the grey fuzzy estimation model were compared and analyzed with multiple linear regression,support vector machine,BP neural network and random forest to verify the validity of the model.The results show that the estimation accuracy of the two methods,grey fuzzy estimation model and self-feedback grey fuzzy estimation model,is higher,in which the determination coefficients R~2of the 20 samples to be identified are 0.9213 and 0.9408,and the average relative errors are6.3630%and 6.9717%,respectively.And the average relative errors of the four estimation models,namely,multiple linear regression,support vector machine,BP neural network,and random forest,were 12.3947%,13.9689%,13.6411%,and 9.6269%,respectively,with determination coefficients R~2of 0.8102,0.7460,0.7384,and 0.8335,respectively.The study shows that the grey fuzzy estimation model can effectively deal with the three uncertainties in spectral estimation and improve the accuracy of soil organic matter spectral estimation. |
