Osmotic Treatment of Fish in Two Different Solutions‐Artificial Neural Network Model

Osmotic treatment of fish (Carassius gibelio) was studied in two osmotic solutions (ternary aqueous solution – S₁ and sugar beet molasses – S₂) at three temperatures (20, 35 and 50C) and atmospheric pressure. The aim was to examine the influence of type and concentration of used hypertonic agent, temperature and immersion time on the water loss, solid gain, dry mater content, water activity and minerals content (Na, K, Ca and Mg). During experiments, the maximum mineral content has been obtained using S₂ solution, concentrated to 80%, at 50C after 5 h of osmotic treatment, at which maximum water loss has been obtained. Artificial neural networks (ANN) have been developed for mathematical modeling of observed responses, and afterwards they were compared with experimental results and empirical linear multivariate regression models. ANN models performed high prediction accuracy (0.975–0.993) and can be considered as precise and very useful for outputs production. PRACTICAL APPLICATIONS: The osmotic treatment (OT) of food is commonly used technique for food processing, mostly utilized prior to drying and freezing operations, which reduces energy requirements of these processes. This study investigates the OT of fish (Carassius gibelio), in two hypertonic solutions (ternary aqueous solution and sugar beet molasses) at atmospheric pressure. The influence of hypertonic agent concentration, temperature and immersion time on the water loss, solid gain, dry mater content, water activity and minerals content (Na, K, Ca and Mg) were studied. Developed artificial neural network mathematical models performed high prediction accuracy: 0.975–0.993 and can be considered as precise for process parameters prediction and optimization in experimental and industrial applications. The wide range of processing variables were considered in the model formulation, and its easy implementation in a spreadsheet using a set of equations makes it very useful and practical for outputs prediction.