Related Experiment Video
Updated: Aug 5, 2026
07:34
Excitation-Scanning Hyperspectral Imaging Microscopy to Efficiently Discriminate Fluorescence Signals
Published on: August 22, 2019
Quantification of individual solid wastes from a mixture using hyperspectral imaging and machine learning
1State Key Laboratory of Coal Combustion, School of Energy and Power Engineering, Huazhong University of Science and Technology, Wuhan 430074, PR China.
Summary
This study introduces a non-destructive method using near-infrared hyperspectral imaging and machine learning to quantify components in mixed solid waste, enabling precise waste blending and utilization.
Area of Science:
- Waste Management & Recycling
- Analytical Chemistry
- Materials Science
Background:
- Accurate quantification of mixed solid waste components is crucial for effective waste management and resource recovery.
- Current methods may be destructive or lack the precision needed for complex waste streams.
Purpose of the Study:
- To develop a non-destructive quantitative framework for analyzing five-component mixed solid waste.
- To benchmark various machine learning models for hyperspectral data analysis in waste quantification.
Main Methods:
- Coupling near-infrared (NIR) hyperspectral imaging with machine learning (ML).
- Utilizing Standard Normal Variate (SNV) preprocessing and Successive Projections Algorithm (SPA) for feature selection.
- Benchmarking five ML models: PLSR, XGBoost, SVR, RF, and 1D-CNN.
Main Results:
- The 1D-CNN model achieved high accuracy (R² = 0.977-0.989) across all components.
- SHAP analysis provided interpretability of spectral features and component interactions.
- The reinforced model showed excellent stability and transferability in extrapolation tests.
Conclusions:
- The developed framework offers a scalable and interpretable approach for rapid quantitative analysis of mixed solid waste.
- This technology has significant potential for precision blending and downstream utilization of waste materials.
