Related Experiment Video
Updated: Aug 5, 2026
08:49
Multimodal Nonlinear Hyperspectral Chemical Imaging Using Line-Scanning Vibrational Sum-Frequency Generation Microscopy
Published on: December 1, 2023
Explainable Deep-Shallow Feature Fusion of Two-Dimensional Encoded Vis-NIR Spectra and RGB Image Features for Chilled
Yanjie Ren1,2, Qi Zhang2,3, Yongqian Zhou1,2
1College of Information Science and Technology, Shihezi University, Shihezi 832003, China.
Foods (Basel, Switzerland)
|July 28, 2026
Summary
This study introduces a new method for classifying chilled lamb freshness by combining visual and spectral data. The integrated approach accurately assesses lamb quality, improving meat safety and control.
Area of Science:
- Food Science
- Spectroscopy
- Computer Vision
Background:
- Chilled lamb quality deterioration necessitates rapid and accurate freshness assessment.
- Current methods using only spectral or RGB image data are insufficient for comprehensive quality evaluation.
- Simultaneously characterizing internal chemical and external appearance changes is crucial.
Purpose of the Study:
- To develop an integrated method for chilled lamb freshness-grade classification.
- To combine deep features from 2D spectral encoding with RGB image features.
- To improve the accuracy and efficiency of lamb quality assessment.
Main Methods:
- Transformed 1D Vis-NIR spectra into 2D images (GADF, GASF, MTF, RP) to capture spectral relationships.
- Fused spectral deep features (selected by RFE) with RGB image features (selected by Spearman).
- Constructed and evaluated a deep-shallow classification model.
Main Results:
- Fusion models significantly outperformed single-modality models.
- The GADF(10%)+Image-SVM model achieved the highest accuracy (0.966), F1-score (0.957), and MCC (0.946).
- SHAP analysis confirmed GADF deep features as primary contributors, with RGB features offering complementary information.
Conclusions:
- The proposed integrated method demonstrates high potential for rapid, non-destructive freshness-grade classification of chilled lamb.
- Combining spectral deep features and RGB image features enhances classification performance.
- This approach offers a robust solution for meat quality and safety control.