Bulletin of Surveying and Mapping ›› 2025, Vol. 0 ›› Issue (5): 27-33.doi: 10.13474/j.cnki.11-2246.2025.0505

Previous Articles    

Hyperspectral remote sensing inversion of zinc content in farmland around metal smelter

GUO Bin1, ZHANG Liye1, ZOU Bin2, GUO Xianan1, BAI Haorui1, ZHANG Bo2, GUO Tengyue3, WU Min1   

  1. 1. College of Geomatics, Xi'an University of Science and Technology, Xi'an 710054, China;
    2. School of Geosciences and Info-physics, Central South University, Changsha 410083, China;
    3. Department of Geological Engineering, Qinghai University, Xining 810016, China
  • Received:2024-10-11 Published:2025-06-05

Abstract: The weak spectral response of trace elements such as soil heavy metals makes feature band extraction challenging,leading to lower accuracy in inversion models. To address these issues,this study intends to use a new approach that extracts spectral response prior knowledge from class-standard soil samples. Firstly,a total of 226 surface(0~20 cm) soil samples were collected from farmland surrounding a zinc smelting plant,with 53 samples used to prepare class-standard soil samples, and both soil zinc content and reflectance spectra were measured. Then,continuous wavelet transform (CWT) was employed to reconstruct the spectral reflectance,and the transfer component analysis (TCA) algorithm was applied to transfer the prior knowledge from class-standard soil samples to ZY-1-02D image spectra. Finally,quantitative inversion models for Zn content were fitted using random forest (RF) and extreme learning machine (ELM),and the optimal inversion model was determined based on accuracy evaluation. The results indicate that transfer learning can achieve spectral response enhancement and improve model accuracy.The paper provides a reference for the study of retrieving relevant soil parameters using prior knowledge.

Key words: heavy metals, ASD FieldSpec4, machine learning, ICP-MS, cross-validation, in-situ sampling

CLC Number: