Study On Hyperspectral Estimation Of Chromium Content In Soil Based On Grey Relational Analysis

Heavy metal content in soil has an important impact on soil environmental quality and crop growth.Therefore,the use of hyperspectral techniques to estimate soil heavy metal content quickly and accurately is critical to soil ecological environment and food safety.In this paper,the Zhaoyuan,Longkou and Qixia regions of Yantai city were taken as research areas,and 70 soil samples collected were taken as research objects.Based on the soil spectral data collected indoors and the content of heavy metal chromium in laboratory tests,statistical analysis,grey correlation analysis and other methods were used to quantitatively estimate the heavy metal content of soil using hyperspectral techniques.The main research contents and results are as follows:(1)Spectral characteristics of soil reflectance and pollution degree of heavy metals were analyzed.The pollution degree of heavy metals in soil was analyzed by the method of geoaccumulation index,and the spectral characteristics of grouped spectra,pollution degree and different soil types were analyzed.The results showed that the heavy metals such as Cr(Chromium),Ni(Nickel),Zn(Zinc)and As(Arsenic)were not polluted,while the heavy metals such as Cu(Copper),Pb(Lead)and Cd(Cadmium)were mild to moderate,and Hg(Mercury)was moderate.Soil spectral reflectance decreases with the increase of chromium content,showing a negative correlation.Meanwhile,soil spectral reflectance decreases with the increase of soil pollution,which is consistent with the former.The overall segregation of the four soil types is not obvious,so it is difficult to distinguish soil types according to soil reflectance,which needs to be determined by combining other factors.(2)The sensitive bands and characteristic factors of chromium content in soil were determined.The original spectral reflectance of soil was transformed by nine transformation methods,such as logarithm,reciprocal,square root,first-order differential and reciprocal first-order differential.The sensitive bands and characteristic factors of chromium content in soil were determined by analyzing the correlation between spectral transformation value and chromium content.The results showed that the effect of the reciprocal,logarithmic and square root transformation of reflectance was not obvious,but the differential transformation can effectively improve the correlation of some bands.Therefore,based on the principle of maximum correlation,five characteristic factors were selected in the three transformation methods: 1910.5 nm of first-order differential transformation,674.1 nm of first-order logarithmic reciprocal differential transformation,1609.4 nm of first-order differential transformation and 1231.3 nm and 1127.3 nm of first-order reciprocal differential transformation.The highest correlation was about 0.75.After transformation,the correlation was improved in varying degrees.(3)A hyperspectral grey correlation estimation model for heavy metal chromium content in soil was established.Based on the selected characteristic factors,the multi-linear regression estimation method based on grey correlation analysis method and several common estimation models,such as multivariate linear regression,BP neural network,grey correlation degree,grey correlation degree correction model,were used to quantitatively estimate the content of heavy metal chromium in soil.The results showed that the accuracy of the multivariate linear regression method based on grey correlation analysis was the best.According to the principle of maximal relevance,when the corresponding pattern selected for each sample was 9 or 12,the estimation effect were better.The model determination coefficients were 0.945 and 0.975,and the average relative errors were 9.901% and 10.919%,respectively.The second was the correction model of grey relation degree.The decision coefficient of this model was 0.968 and the average relative error was 11.518%.The accuracy of the model was higher than that of other common models.The results showed that the method of multiple linear regression estimation based on grey correlation analysis is feasible and effective for quantitative estimation of soil heavy metal chromium content.