基于SPA-BPNN的成都市天府新区东部土壤As含量高光谱估测建模OACSTPCD

The problem of soil pollution by heavy metals has become increasingly severe,which is threatening the ecological environment and human health.In order to clarify the heavy metal As in soils in the urban fringe zone of Chengdu plain,this study takes the east of Tianfu New District in Chengdu,Sichuan as the research ob-ject to perform the soil raw spectral data by first-order differential(FD),second-order differential(SD),de-en-velope processing(CR),standard normal transform(SNV).The Pearson correlation coefficient(PCC)and suc-cessive projections algorithm(SPA)are used to filter the characteristic bands of the best transformed spectra.Four regression models,namely,partial least squares(PLSR),extreme learning machine(ELM),random for-est(RF)and BP neural network(BPNN),are developed for the estimation of soil heavy metal content based on the hyperspectral data,and then validate the accuracy.The results show that the correlation between the spectra and the As content of soil is significantly improved by de-enveloping first-order differential(CR-FD)transforma-tion,from 0.473 to 0.848;the non-linear model is higher than the linear model in terms of model fit and pre-diction model accuracy,whether based on the PCC or SPA algorithm to extract the feature bands.Compared with the modelling results based on the PCC algorithm,the prediction accuracy of the models built with the fea-ture variables filtered by the SPA algorithm is significantly improved,with R2 of 0.786,0.847,0.856 and 0.942 for the PLSR,ELM,RE and BPNN model validation sets respectively.The BPNN model,which is con-structed using the spectral bands selected by the SPA as the independent variables,gives the best results for the estimation of the heavy metal content in the study area.