Jiangsu Province Key Laboratory of Live Food for Fisheries, School of Marine Science and Engineering, Nanjing Normal University, No. 2 Xuelin Rd, Nanjing, 210023, People's Republic of China
Auto-generated profile, based on publicly available information
Cui Han
Jiangsu Province Key Laboratory of Live Food for Fisheries, School of Marine Science and Engineering, Nanjing Normal University, No. 2 Xuelin Rd, Nanjing, 210023, People's Republic of China
The geographical tracing of high-value salmonids is essential in mitigating food fraud and ensuring authenticity. This study combined hyperspectral imaging with elemental/isotopic profiling to reflect geographical origin information of salmonids and improve the explainability of discrimination models. Pearson correlation analysis was conducted using spectra and elements/isotopes for spectral fingerprinting, followed by developing machine/deep learning models with full spectra and extracted spectral features, respectively. The key isotopes and elements were further predicted using random forest modeling. The results indicated elements and isotopes highly correlating with spectral regions around 900, 1000, and 1200 nm, reflecting important spectral profiling of geographical origins for fresh salmonid products. Further, all discrimination models achieved acceptable overall accuracies (>65 %), with SFS-ELM achieved the highest precision (88.13 %). Key elements/isotopes were satisfactorily predicted (R2 > 0.70) with chemical maps for vivid visualization. This study combined HSI with elemental/isotopic analysis to provide a chemometric basis for interpreting spectral regions that are most relevant to origin discrimination, improving the comprehensive understanding of model interpretability and feature validation.