Prediction of Vanadium Contamination Distribution Pattern Through Remote Sensing Image Fusion and Machine Learning DOI Creative Commons

Z. G. Zhao,

Yuman Sun, Weiwei Jia

и другие.

Remote Sensing, Год журнала: 2025, Номер 17(7), С. 1164 - 1164

Опубликована: Март 25, 2025

Soil vanadium contamination poses a significant threat to ecosystems. Hyperspectral remote sensing plays critical role in extracting spectral features of heavy metal contamination, mapping its spatial distribution, and monitoring trends over time. This study targets vanadium-contaminated area Panzhihua City, Sichuan Province. sampling measurements occurred the laboratory. (Gaofen-5, GF-5) multispectral (Gaofen-2, GF-2; Sentinel-2) images were acquired preprocessed, feature bands extracted by combining laboratory data. A dual-branch convolutional neural network (DB-CNN) fused hyperspectral confirmed fusion’s effectiveness. Six prevalent machine learning models adopted, unified framework leveraged Random Forest (RF) as second-layer model enhance predictive performance these base models. Both ensemble evaluated based on accuracy. The fusion process enhanced models, improving R2 values for (V) pentavalent (V5+) from 0.54 0.3 0.58 0.39, respectively, at 4 m resolution. Further optimization using RF refine Extreme Trees (ETs) significantly increased 0.83 0.75 V V5+, this scale. 934 nm 464 wavelengths identified most predicting soil contamination. integrated approach robustly delineates distribution characteristics V5+ soils, facilitating precise ecological risk assessments through comparative analysis accuracy across diverse

Язык: Английский

Prediction of Vanadium Contamination Distribution Pattern Through Remote Sensing Image Fusion and Machine Learning DOI Creative Commons

Z. G. Zhao,

Yuman Sun, Weiwei Jia

и другие.

Remote Sensing, Год журнала: 2025, Номер 17(7), С. 1164 - 1164

Опубликована: Март 25, 2025

Soil vanadium contamination poses a significant threat to ecosystems. Hyperspectral remote sensing plays critical role in extracting spectral features of heavy metal contamination, mapping its spatial distribution, and monitoring trends over time. This study targets vanadium-contaminated area Panzhihua City, Sichuan Province. sampling measurements occurred the laboratory. (Gaofen-5, GF-5) multispectral (Gaofen-2, GF-2; Sentinel-2) images were acquired preprocessed, feature bands extracted by combining laboratory data. A dual-branch convolutional neural network (DB-CNN) fused hyperspectral confirmed fusion’s effectiveness. Six prevalent machine learning models adopted, unified framework leveraged Random Forest (RF) as second-layer model enhance predictive performance these base models. Both ensemble evaluated based on accuracy. The fusion process enhanced models, improving R2 values for (V) pentavalent (V5+) from 0.54 0.3 0.58 0.39, respectively, at 4 m resolution. Further optimization using RF refine Extreme Trees (ETs) significantly increased 0.83 0.75 V V5+, this scale. 934 nm 464 wavelengths identified most predicting soil contamination. integrated approach robustly delineates distribution characteristics V5+ soils, facilitating precise ecological risk assessments through comparative analysis accuracy across diverse

Язык: Английский

Процитировано

0