Bayesian Fusion Model Enhanced Codfish Classification Using Near Infrared and Raman Spectrum

Area of Science:

• Food Science
• Analytical Chemistry
• Spectroscopy

Background:

• Accurate and rapid identification of fish species is crucial for food quality control and preventing fraud.
• Traditional methods for fish speciation can be time-consuming and destructive.
• Spectroscopic techniques offer potential for non-destructive analysis, but often require advanced data processing for optimal performance.

Purpose of the Study:

• To develop and evaluate a novel Bayesian-based decision fusion technique for the identification of codfish.
• To compare the performance of the Bayesian fusion model with individual near-infrared (NIRS) and Raman spectroscopy (RS) methods, as well as conventional data and feature layer fusion approaches.
• To assess the efficacy of combining NIRS and RS data for enhanced codfish classification.

Main Methods:

• Collection of NIRS and RS spectra from 320 codfish samples.
• Development of separate partial least squares discriminant analysis (PLS-DA) models for NIRS and RS data.
• Implementation and comparison of three decision fusion methods: conventional data layer fusion, feature layer fusion, and a novel Bayesian-based decision fusion (NIRS-RS-B).
• Extraction of optimal discrimination features from NIRS and RS data for the Bayesian fusion model.

Main Results:

• The Bayesian-based decision fusion model (NIRS-RS-B) achieved superior classification metrics: 92% sensitivity, 98% specificity, and 98% accuracy.
• The developed Bayesian fusion approach significantly outperformed individual NIRS and RS methods.
• The Bayesian model also demonstrated significantly better performance compared to conventional data layer (NIRS-RS-D) and feature layer (NIRS-RS-F) fusion methods.
• The PLS-DA models effectively established relationships between spectral data and codfish identity.

Conclusions:

• The novel Bayesian-based decision fusion technique integrating NIRS and RS provides a highly accurate and efficient method for non-destructive codfish identification.
• This approach offers a significant advancement over existing spectroscopic and fusion methods for food speciation.
• The developed technique holds potential for application in identifying other food species and in various food authentication scenarios.