Freshness assessment of olive flounder fillets using hyperspectral imaging with data fusion strategies

We aimed to evaluate the freshness of olive flounder ( Paralichthys olivaceus ) fillets stored under modified atmosphere packaging at 5, 10, 15, and 20 °C. Microbiological counts, total volatile bases, color, and free amino acids were measured, and samples were classified into three freshness categories (fresh, acceptable, and spoiled) based on quality parameters. Hyperspectral images were acquired in the visible and near-infrared (VNIR) and short-wave infrared (SWIR) ranges, and support vector machine models were applied for nondestructive freshness classification. Models developed from single spectra achieved accuracies of 86.2 % (VNIR) and 75.9 % (SWIR). To enhance predictive performance, the two spectral ranges were fused through low-level fusion (LLF) and mid-level fusion (MLF) incorporating agglomerative clustering (AC). Classification models incorporating LLF and MLF outperformed the single-spectra models. Notably, MLF2, which applied AC to the LLF dataset, achieved an accuracy of 94.8 % while using only 28 % of the full wavelength range. These findings demonstrate that appropriate fusion of complementary spectral information can provide an efficient and reliable approach for evaluating the freshness of fish products. • Hyperspectral imaging enables nondestructive evaluation of olive flounder freshness. • VNIR had a dominant impact on freshness prediction. • Fusion with VNIR and SWIR further improved prediction accuracy. • Data fusion sequence affected prediction performance.