Hyperspectral Inversion Of Heavy Metal Content In Soil Based On Particle Swarm Optimization-back Propagation Neural Network Method

With the fast increase of industrial and chemical pesticide pollutants,many heavy metals enter the soil environment through atmospheric deposition and sewage irrigation.The toxicity of heavy metals is not easily degraded by microorganisms.It is easy to accumulate in animals,plants and humans through the food chain,which induces direct or indirect harms to the environment and humanity.So it is of great significance to seek a rapid and non-destructive monitoring method for soil heavy metal pollution.However,the development of remote sensing technology,especially hyperspectral remote sensing technology,provides a new way to achieve large-scale and rapid monitoring of heavy metal pollution in soil.In view of the low precision and poor universality of soil heavy metal hyperspectral models constructed by statistical analysis and machine learning,in this study,a back propagation neural network（BPNN）was combined with the particle swarm optimization（PSO）,which led to an integrated PSO-BPNN method used to estimate the contents of soil heavy metals:Cd,Hg,and As.This study was conducted in Guangdong of China based on the soil heavy metal contents and hyperspectral data collected from 95 soil samples.firstly,the content of three heavy metal elements（Cd,Hg,As）in the soil samples was determined by indoor chemical analysis method,and the spectral data of the soil samples were collected by means of the ground object spectrometer;Secondly,the original spectrum is spectrally transformed and double-band combined,and the optimal band factor combination variable is extracted by correlation analysis and variance expansion factor（VIF）.Thirdly,the BPNN and PSO-BPNN methods are used to establish the soil heavy metal spectrum estimation model,and the models constructed by the two methods are validated and compared.Finally,the optimal estimation model is combined with the environmental remote sensing image to carry out the soil heavy metal content.Inversion,the spatial distribution of Cd content of heavy metals in the soil of the study area was obtained.The results showed that:1)The sample averages of Cd,Hg and As were 0.174mg·kg^-1,0.132 mg·kg^-1 and 9.761 mg·kg^-1,respectively,with the corresponding maximum values of 0.570 mg·kg^-1,0.310 mg·kg^-1 and 68.600 mg·kg^-1 being higher than the environment baseline values.2)The transformed and combined spectral variables had higher correlations with the contents of the soil heavy metals than the original spectral data.Among them,the First Derivative and the Continuum Removal transformation forms have a significant impact on the correlation,which is an effective spectral index;however,the band combination method can significantly improve the correlation between spectral variables and heavy metal content,correlation coefficient both reached 0.42 or more.3)In this paper,the VIF analysis method can eliminate the problem of multicollinearity and information redundancy between spectral variables,so as to obtain the best spectral index combination variables of soil heavy metals Cd,Hg and As.They are FD_938.753*FD_795.231,LG784.504/LG492.442;FD_1373.48+7*FD_430.21,LG2222.424-LG1212.22,7*RT2222.424-12*RT1212.22;FD2342.058/FD966.869,RT343.281/RT343.874,6*SD_363.425-5*SD_340.316.It is found that the original spectral bands do not appear in the optimal spectral band combination of heavy metal elements in each soil,indicating that the effect of using the spectrally transformed characteristic band to participate in the model construction is better than the estimation using the original spectrum.4)PSO-BPNN significantly improved the estimation accuracy of the soil heavy metal contents with the decreases of mean relative error（MRE）and relative root mean square error（RRMSE）by 68%to 71%and 64%to 67%,respectively.This indicated that PSO-BPNN provided great potential to estimate the soil heavy metal contents.5)With PSO-BPNN,the Cd content could also be mapped using Huan Jing-1A Hyperspectral Imager（HSI）data with a RRMSE value of 36%,implying that the PSO-BPNN method could be utilized to map the heavy metal content in soil using both field spectral data and hyperspectral imagery for the large area.