Application of Hyperspectral Imaging and Artificial Intelligence for Freshness Assessment of Fish Raw Material during Refrigerated Storage

Abstract

Introduction: Monitoring the freshness of refrigerated fish remains one of the persistent difficulties in the fish processing industry. Existing reference methods for assessing refrigerated fish freshness are destructive and inherently unable to reflect the spatial distribution of spoilage markers. Hyperspectral imaging (HSI) has emerged as a powerful non-destructive tool that captures both spatial and spectral data at the pixel scale. Purpose: The aim of this study is to evaluate the effectiveness of hyperspectral imaging for the binary classification of refrigerated rainbow trout (Oncorhynchus mykiss) fillets into early (≤48 h) and late (>48 h) storage stages. Materials and Methods: The study was conducted on rainbow trout fillets stored at +2 ± 2 °C for 16 days. Hyperspectral images were acquired using a FigSpec FS-23 camera (spectral range 400–1000 nm). Principal component analysis (PCA) was performed, and a neural network model for binary classification of samples was developed using the TensorFlow framework and the high-level Keras API. Results: A characteristic nonlinear dynamics of reflectance was observed during storage. PCA showed that the first principal component accounted for 93.8 % of the data variance. The neural network model achieved 90 % accuracy in binary classification of the samples. Conclusion: The results demonstrate the potential of hyperspectral imaging as a non-destructive tool for assessing fish freshness. The developed method provides accurate discrimination between fresh and non-fresh samples and can be recommended for adoption in industrial incoming inspection protocols.