Physicochemical properties and gel quality monitoring of surimi during thermal processing using hyperspectral imaging combined with deep learning

The fish processing industry faces significant challenges in maintaining the quality of surimi products, which can impact consumer appeal and overall product quality. This study investigated the use of hyperspectral imaging for rapid, non-destructive, online monitoring of quality changes during two-stage water bath heating of surimi. Surimi samples were heated at 40 °C for 30 min and 90 °C for 20 min, with data collected every 5 min on key quality indicators, including gel strength, water-holding capacity, and whiteness. To comprehensively evaluate the quality changes of surimi during the two-stage heating process, two models were developed using hyperspectral imaging: a partial least squares (PLS) model and a convolutional neural network-long and short-term memory (CNN-LSTM) model. A separate CNN-LSTM model was created to predict multiple quality indicators simultaneously. Although the optimal model for predicting individual quality indicators slightly outperformed the multi-indicator predicting model ($R_p^2$ > 0.93, RPD >3.9), both approaches were effective. Additionally, the quality changes observed during the heating process were visualized and analyzed. This study demonstrates that hyperspectral imaging, combined with chemometrics, offers a non-destructive, rapid, and online method for predicting quality changes during the thermal processing of surimi. This approach addresses the industry's need for innovative quality assessment tools and has the potential to enhance product quality and consumer satisfaction in the processed surimi market.