Real-Time Epizootic Monitoring with Inception Deep Neural Network for Maritime Applications

Abstract

This study explores the integration of artificial intelligence in aquaculture to differentiate healthy fish from those afflicted with diseases, aiming to establish a real-time, automated epizootic monitoring system. Utilizing the “Inception v3” convolutional neural network, we examined the model's efficacy in classifying fish based on their health status across two experiments focusing on data augmentation variability. Initial results without augmentation showed diseased fish detection with 86.7% accuracy and healthy fish detection with 86.9% accuracy. However, employing diverse augmentation techniques significantly enhanced detection accuracy to 96.9% for diseased fish and 96.7% for healthy specimens. These findings underscore the potential of computer vision technologies in revolutionizing epizootic monitoring in aquaculture by providing a non-invasive, accurate, and scalable solution to fish health management. The successful application of AI in this context could significantly contribute to the sustainability and productivity of aquaculture operations, underscoring a pivotal shift towards more advanced and humane practices in the industry.