Parallel Mirrors Based Marine Predator Optimization Algorithm with Deep Learning Model for Quality and Shelf-Life Prediction of Shrimp

Abstract

Automatic classification and assessment of shrimp freshness plays a major role in aquaculture industry. Shrimp is one of the highly perishable seafood, because of its flavor and excellent nutritional content. Given the high amount of industrial production, determining the freshness of shrimp quickly and precisely is difficult. Instead of using feature-engineering-based techniques, a novel hybrid classification approach is proposed by combining the strength of convolutional neural networks (CNN) and Marine Predators Algorithm (MPA) for shrimp freshness diagnosis. In order to choose the best hyperparameter values, marine predator algorithm is improved using Parallel Mirrors Technique (PMPA). The proposed methodology employs a pretrained CNN model termed EfficientNet (ENet), which is combined with the PMPA algorithm to form the PMPA-ENet architecture. The proposed approach yields high performance while also reducing computational complexity. The result showed that proposed model achieved an accuracy and F-score of 98.62% and 97.25% for assessment of freshness in shrimp. PMPA's effectiveness in determining optimal values is compared to four different meta-heuristic algorithms hybridized with ENet including Particle Swarm Optimization (PSO), Simple Genetic Algorithm (SGA), Whale Optimization Algorithm (WOA), and traditional Marine Predator Algorithm (MPA). The results indicated that PMPA-ENet algorithm provides better classification compared with other algorithms