An Ensemble Based Deep Learning Architecture for Fish Disease Detection and Classification

Abstract

Fish diseases pose a major threat to aquaculture, leading to significant economic and ecological consequences. Fish diseases are a serious problem in aquaculture, causing economic losses and harming the environment. This study presents an ensemble deep learning model that combines ResNet50 and EfficientNetB0 to accurately detect and classify fish diseases from images. A total of 1,747 images were collected and grouped into seven classes, including different types of bacterial, fungal, viral, parasitic diseases, and healthy fish. Several preprocessing steps were used to improve image quality and help the model perform better. The ensemble model uses the strengths of both networks, along with global average pooling and dense layers, to make predictions. The model performed very well, achieving 98.5% accuracy, 98% precision, 97.9% recall, and a 97.8% F1-score. The confusion matrix, training and validation accuracy, and training and validation loss show that the model is reliable and works well across different disease types. These results suggest the model can be used in real-time systems to monitor fish health, reduce disease-related losses, and support food security. This study shows that using ensemble deep learning can be a powerful, accurate, and practical solution for managing fish diseases in aquaculture, which can significantly benefit aquaculture productivity, sustainability, and food security.