An Enhanced Aquaculture Health Monitoring System For Fish Veterinarians For Identifying And Differentiating Different Types Of Freshwater Fish Diseases

Abstract

In Aquatic Habitats, the fish population's viability is pivotal for preserving ecosystem stability and enhancing aquaculture productivity. Early and precise identification of fish diseases is of utmost importance for strategic management and risk reduction approaches. This investigation explores a robust approach for fish health assessment using transfer learning and pre-trained neural network architectures. The proposed methodology uses encoded representations derived from ten Fine-Tunable Pre-Trained Models, namely VGG-16, VGG19, ResNet50, MobileNet V2, and Inception-V3, for designing a trained model for classification. Through rigorous experimental analysis using an extensive dataset, we have highlighted the importance of our suggested model designed using VGG16 in precisely recognizing seven different types of fish diseases namely Disease-free, fungal infection, parasitic infestation, and aquatic viral infections. The trained model designed using VGG16 outperformed other models in overall classification accuracy (0.97857), Positive Predictive Value (0.97901), True Positive Rate (0.97857), and Kappa (0.975) respectively. Our revelations expose the emerging strengths of transfer learning and customizing the layers for strengthening fish disease surveillance and classification, unlocking new opportunities for emerging research directions in aquatic health monitoring and aquaculture.