Abstract: The intersection of Qingyuan County, Anxin County and Gaoyang County in Baoding City used to be an iron and steel smelting and processing area. Due to years of extensive smelting, there is a certain amount of soil heavy metal pollution in this area. Firstly, takes the junction of three counties as the study area, obtains the basic data of the study area through soil sample collection, soil spectral measurement and laboratory content determination of lead and zinc elements. Then, carries out correlation analysis, statistical analysis and model modeling inversion on the basic data. Finally, two modeling methods, partial least squares and back propagation neural network are used to establish the quantitative estimation model of content of lead and zinc elements. The results show that six mathematical transformations are beneficial to establish the characteristic bands of correlation between hyperspectral bands and content of lead and zinc elements by Pearson correlation analysis; based on the analysis of six spectral transformations, partial least squares and back propagation neural network, the back propagation neural network can improve the prediction accuracy of content of lead and zinc elements in soil; the estimation model of lead element based on continuum removal established by back propagation neural network is the best estimation model, in which the modeling accuracy R² is 0.949 and the verification accuracy R² is 0.980; the estimation model of zinc element based on standard transformation established by back propagation neural network is the best estimation model, in which the modeling accuracy R² is 0.874 and the verification accuracy R² is 0.957. Through the study, the estimation model established by the back propagation neural network can provide a research basis for the investigation of soil heavy metal lead and zinc pollution in this area.