文献类型： 外文期刊

作者： Chen, Xuying ¹ ; Li, Long ¹ ; Li, Xuqing ¹ ; Ou, Huiping ² ; Liu, Tianjiao ⁴ ; Gu, Huitao ¹ ; Wang, Tingxuan ¹ ;

作者机构： 1.North China Inst Aerosp Engn, Sch Remote Sensing & Informat Engn, Langfang 065000, Peoples R China

2.Guangxi Acad Agr Sci, Agr Resources & Environm Res Inst, Guangxi Key Lab Arable Land Conservat, Nanning 530007, Peoples R China

3.Beijing Normal Univ, State Key Lab Remote Sensing Sci, Beijing, Peoples R China

4.Chinese Acad Agr Sci, Inst Agr Resources & Reg Planning, State Key Lab Efficient Utilizat Arable Land China, Beijing, Peoples R China

关键词： Hyperspectral imagery; farmland; soil heavy metals; band screening; Stacking

期刊名称：INTERNATIONAL JOURNAL OF REMOTE SENSING （ 影响因子：2.6； 五年影响因子：2.9 ）

ISSN： 0143-1161

年卷期： 2025 年 46 卷 17 期

页码：

收录情况： SCI

摘要： Human health and food security are seriously threatened by heavy metal pollution of farming soil. Consequently, to timely and accurately determine the heavy metal concentration of agriculture soil, efficient monitoring techniques needs to be put in place. Traditional monitoring methods often prove inadequate in meeting the demands of large-scale and high-timeliness monitoring. This study proposes a Stacked-PLSR inversion model for the assessment of heavy metal content in agricultural soils. Using ZY-01-02D hyperspectral image data, the model inverts the Cu, As, and Ni contents in soil using eXtreme Gradient Boosting(XGBoost), Random Forest(RF), and Multilayer Perceptron(MLP) as the base model and Partial Least Squares Regression(PLSR) as the metal model. The study employs logarithmic and inverse re-first-order differential processing to enhance the spectral features and combines the Particle Swarm Optimization(PSO) algorithm with the Pearson correlation coefficient(PCC) for feature optimization. The experimental findings demonstrate that the Stacked-PLSR model outperforms the conventional machine learning model with respect to stability and prediction accuracy. The validation set R2 values for Cu, As, and Ni are 0.890, 0.724, and 0.844, respectively.Based on the model's inversion results, the heavy metals' spatial distribution map shows that the research area's agricultural soils are largely uncontaminated. However, there is a notable elevated concentration around the industrial area, indicating that the industrial area may serve as a potential source of pollution. This finding suggests that the industrial area may be a contributing factor to the observed levels of pollution.A new technical method for high-precision, large-scale remote sensing monitoring of heavy metal pollution in agricultural soils is proposed in this work. This approach is of significant importance in achieving precise pollution prevention and control, and ensuring food and ecological environment safety.