| The near-surface soil hyperspectral analysis technology has been widely used cause its convenience and rapidity,which can obtain soil information in a large range,and gradually become an important means of dynamic monitoring of key index of soil fertility.In the process of field spectral measurement,it is easy to be affected by environmental light,such as soil moisture,texture,plant litter and other factors.Among them,the color of surface plant litter and its morphological characteristics of mixed spectrum are similar to the soil,which lead to ignore the influence on the soil spectrum during the observation process easily,and reduce the accuracy of the soil spectral estimation model.In order to further study the effect of plant litter layer on the hyperspectral estimation of soil organic carbon and an effective method of unmixing,this paper takes the soil spectrum of Hulunbeier’s typical grassland as the research object.Using HR-1024 spectrometer to obtain indoor simulation and field soil spectra under different coverage,and analysis its spectral characteristics and regularity.The mixed spectrum is separated by infomax algorithm,and the validity and accuracy of infomax algorithm are further verified by using partial least square(PLSR)and support vector machine(SVM)to model and predict the organic carbon content before and after unmixing.Then,the algorithm is applied to the field measurement of spectral unmixing to verify its applicability.The results show that:1)Based on the analysis of the spectral characteristics of soil reflectance under different coverage,it is concluded that the cellulose absorption index(CAI)increases gradually with the coverage of litter increases on the earth’s surface,also shows a secondary increasing relationship with the coverage of litter on the earth’s surface.In the presence of litter coverage,the spectral reflectance increases in the whole band,the spectral slope increases sharply at 700nm,the lignin absorption at 1680nm and 1754nm is weak,and the cellulose absorption band at 2100nm gradually expands and deepens.The absorption characteristics of lignin and cellulose in the field are more obvious than that in the laboratory.2)Through the analysis of the influence of soil mixing spectrum on the estimation of organic carbon,it is concluded that when the content of soil organic carbon is large,the absorption band at 2100nm is higher than the curve of bare soil,and the boundary is clear,and the absorption band area is wide.Therefore,it can be inferred that when the organic carbon content is high,the soil spectrum is easily affected by litter,and the characteristics of litter absorption zone are significant.The correlation coefficient between the mixed spectral data and the organic carbon under different coverage is calculated.According to the result,the correlation will be reduced when there is litter coverage.When using bare soil spectral modeling and different coverage data as prediction set to estimate the organic carbon content,the organic carbon content is significantly overestimated.Even if only 5%of the soil surface is covered with litter,the organic carbon content will be overestimated by 11.94%.3)It is found that the best result can be obtained when the learning step is 0.002 by studying the unmixing effect of Infomax on the mixed spectrum.The spectral similarity value(SSV)is used to determine the accuracy of the solution,and the result shows that the coverage is less than 15%.This algorithm can get better results of the solution.The results showed that the validation set R2 of PLSR model increased from 0.75 to 0.83,and RMSE decreased from 4.48g/kg to 3.98g/kg.The results show that the combination of Infomax and PLSR can improve the precision of soil litter mixed spectrum based on hyperspectral technology. |
PDF Full Text Request