Non-Destructive Detection Pilot Study of Vegetable Organic Residues Using VNIR Hyperspectral Imaging and Deep...

Area of Science:

• Food Science
• Spectroscopy
• Machine Learning

Background:

• Food contamination poses significant risks to public health and necessitates robust safety measures.
• Effective detection and classification of organic residues on food processing equipment are crucial for maintaining hygiene standards.

Purpose of the Study:

• To investigate the efficacy of visible and near-infrared (VNIR) hyperspectral imaging for detecting and classifying organic residues on metallic food processing surfaces.
• To compare the performance of a 1D convolutional neural network (CNN-1D) against other classification methods for residue analysis.

Main Methods:

• Utilized a line-scan VNIR hyperspectral imaging system to acquire data in the 400-1000 nm range.
• Prepared samples by diluting potato and spinach juices to six different concentrations.
• Applied six classification methods, including CNN-1D and five pre-processing techniques, to spectral data.

Main Results:

• The 1D CNN-1D model achieved high classification accuracies: 0.99 (calibration) and 0.98 (validation) for spinach residues, and 0.99 (calibration) and 0.94 (validation) for potato residues.
• CNN-1D outperformed the support vector machine classifier, which showed validation accuracies of 0.90 for spinach and 0.92 for potato.
• Demonstrated the capability of VNIR hyperspectral imaging to differentiate between various concentrations of organic residues.

Conclusions:

• VNIR hyperspectral imaging combined with deep learning, specifically CNN-1D, offers a promising solution for rapid and non-destructive detection and classification of organic residues.
• This technology can significantly enhance food safety protocols within food processing facilities.
• The study highlights the potential of advanced spectral imaging and machine learning for real-time food quality and safety monitoring.