Related Experiment Video
Updated: Jun 28, 2025
00:07
Excitation-Scanning Hyperspectral Imaging Microscopy to Efficiently Discriminate Fluorescence Signals
Published on: August 22, 2019
8.0K
Sugar detection in adulterated honey using hyper-spectral imaging with stacking generalization method
Madhusudan G Lanjewar1, Kamini G Panchbhai2, Lalchand B Patle3
1School of Physical and Applied Sciences, Goa University, Taleigao Plateau, Goa 403206, India.
Food Chemistry
|April 13, 2024
Summary
This study introduces a novel hybrid method using hyper-spectral imaging and machine learning to accurately detect sugar concentration and types in honey non-invasively. The developed stacking model (STM) demonstrates high precision in predicting sugar levels and classifying honey varieties.
Area of Science:
- Agricultural Science
- Analytical Chemistry
- Spectroscopy
Background:
- Accurate determination of sugar content and honey type is crucial for quality control and authenticity verification.
- Traditional methods for honey analysis can be time-consuming, destructive, or require specialized reagents.
- Developing rapid, non-invasive techniques for honey characterization is an ongoing research objective.
Purpose of the Study:
- To develop and validate a hybrid, automated, and non-invasive approach for detecting sugar concentration and classifying honey types.
- To integrate hyper-spectral imaging with advanced machine learning algorithms for enhanced analytical performance.
- To assess the efficacy of a stacking generalization model (STM) for predicting sugar levels and categorizing honey.
Main Methods:
- A hybrid methodology combining hyper-spectral imaging, Savitzky-Golay (SG) Filter, Principal Components Analysis (PCA), and Machine Learning (ML) classifiers/regressors.
- Utilizing stacking generalization methods to build a robust predictive model (STM).
- Training and testing the model on 32 different sugar concentration levels, six sugar ranges, 11 honey types, and 100% sugar samples.
Main Results:
- The stacking model (STM) achieved a high coefficient of determination (R2) of 0.999 and a low Root Mean Square Error (RMSE) of 0.493 ml (v/v) for predicting sugar concentrations.
- The STM demonstrated excellent classification performance with a Matthews Correlation Coefficient (MCC) and Kappa score of 99.7% for sugar ranges and honey types.
- 10-fold cross-validation confirmed the model's robustness, yielding an average R2 of 0.996 and RMSE of 1.27 ml (v/v) for sugar prediction.
Conclusions:
- The developed hybrid, non-invasive approach effectively detects sugar concentration and classifies honey types with high accuracy.
- The stacking generalization model shows significant promise for automated, real-time quality control and authentication of honey.
- This technique offers a viable alternative to conventional analytical methods, enhancing efficiency and reducing sample manipulation.
