Empirical Evaluation of Deep Learning Techniques for Fish Disease Detection in Aquaculture Systems: A Transfer Learning and Fusion-Based Approach

Abstract

In aquatic environments, the health of fish populations is crucial for maintaining ecological balance and sustaining aquaculture industries. Timely and accurately detecting fish diseases is paramount for effective management and mitigation strategies. This paper presents a novel approach for fish disease detection leveraging transfer learning and ensemble techniques. Our method combines features extracted from three pre-trained deep learning models, namely VGG-16, MobileNet-V2, and Inception-V3, with a Support Vector Machine (SVM) for classification. Through empirical experimental evaluation on a comprehensive dataset, we demonstrate the effectiveness of our proposed model in accurately detecting various fish diseases. The results showcase significant improvements in sensitivity and specificity compared to existing approaches. Additionally, we analyze the impact of different transfer learning strategies and feature fusion techniques on the model’s overall performance. Our findings underscore the potential of transfer learning and ensemble methods in enhancing fish disease detection systems, offering promising avenues for future research in aquatic health management and aquaculture.