Predicting the Freshness of Starch-Coated Snakehead Fish Fillets During Storage Using Hyperspectral Imaging Combined with Transfer Learning

Freshness prediction of starch-coated snakehead fish fillets across different storage times remains challenging due to complex quality deterioration and spectral distribution shifts. In the current research, hyperspectral imaging (HSI) combined with transfer learning (TL) was developed to predict the freshness of starch-coated snakehead fish fillets during short-term refrigerated and long-term frozen storage. The results showed that storage led to texture deterioration, pH increase, TVB-N accumulation, and lipid oxidation, while starch coating effectively delayed quality degradation. Compared with models based only on short-term or long-term data, the domain transfer convolutional neural network (DT-CNN) model improved the robustness of freshness prediction across storage stages. The DT-CNN model based on VIS spectra achieved the best performance for TBA prediction in the starch coating treatment group, with an RP2 of 0.76 and RMSEP of 0.13, and showed strong performance for TVB-N prediction in the starch coating treatment group, with an RP2 of 0.85 and RMSEP of 8.66. These results demonstrate that HSI combined with TL is a promising non-destructive method for freshness evaluation of starch-coated snakehead fish fillets during storage.