Towards Fish Disease Detection using the Vision Transformer Model and Convolutional Neural Network Model

Fish disease detection is important for maintaining the health and sustainability of aquaculture. Basic methods which are used to detect fish diseases depend on manual observation but they are often inefficient and subject to errors. This paper shows a better approach combining Vision Transformer models and Convolutional Neural Networks (CNNs) to reduce these limitations. ResNet50 and VGG16 are used for image-based fish disease detection. This technique increases the power and accuracy of illness categorization by integrating global image analysis with local feature extraction. The ViT-Base Patch 16 model performed exceptionally well, getting a precision of 100%, recall of 98.21%, an F1-score of 99.10% for bacterial diseases, and perfect scores (100%) for parasitic diseases. Our models were validated on a dataset of 2,450 images of seven classes, applying data augmentation methods to overcome the limitations of the shorter dataset size. The potential of deep learning technologies to transform disease management and contribute to the sustainability of aquaculture practices is represented in research results.