Analyzing TVB-N in snakehead by Bayesian-optimized 1D-CNN using molecular vibrational spectroscopic techniques:...

Area of Science:

• Food Science
• Analytical Chemistry
• Spectroscopy

Background:

• Total volatile basic nitrogen (TVB-N) is a critical indicator for fish freshness.
• Accurate and rapid assessment of TVB-N is essential for the seafood industry.

Purpose of the Study:

• To evaluate the efficacy of Near-Infrared (NIR) and Raman spectroscopy for TVB-N detection in snakehead fillets.
• To develop and compare predictive models for TVB-N content using spectroscopic data and data fusion techniques.

Main Methods:

• Feature extraction using Variable Crossover Point Arithmetic - Improved Reduced-Input Vector (VCPA-IRIV) from NIR and Raman spectra.
• Development of Partial Least Squares (PLS) and 1D-Convolutional Neural Network (1D-CNN) models.
• Application of data fusion strategies, including feature-level fusion, with Bayesian optimization.

Main Results:

• Variable Crossover Point Arithmetic - Improved Reduced-Input Vector (VCPA-IRIV) successfully extracted relevant features.
• Feature-level data fusion combined with a Bayesian-optimized 1D-CNN model achieved the highest accuracy.
• Calibration and predictive correlation coefficients for TVB-N reached 0.9677 and 0.9676, respectively.

Conclusions:

• NIR and Raman spectroscopy are effective tools for assessing fish freshness.
• The fusion of NIR and Raman spectroscopy offers a rapid, efficient, and comprehensive method for quantifying fish freshness.
• Advanced chemometric and machine learning approaches enhance the predictive capabilities for TVB-N.