A Deep Convolutional Neural Network for Multi-class Fish Disease Classification using VGGNet-19 Architecture

Background: Accurate fish disease identification is important for ecological research, biodiversity monitoring and conservation. Traditional methods are time-consuming and error-prone. Deep learning provides an automated solution for species classification using image data. Fish disease classification using image-based recognition is an emerging research area, critical for the aquaculture industry. Accurate disease detection is essential for preventing outbreaks, minimizing economic losses and ensuring fish health. Methods: This study employs the VGGNet-19 architecture to detect fish diseases. The dataset, sourced from Kaggle, contains images of fish affected by various diseases. Images are pre-processed with resizing, normalization and data augmentation techniques, including random flipping, rotation and zoom. The model is trained with an 80%-10%-10% split for training, validation and testing. Performance is evaluated using accuracy, a confusion matrix, a classification report and ROC curves. Result: The custom model achieved an overall accuracy of 90.96%. It performed well across most disease categories, correctly distinguishing between healthy and infected fish. Some misclassifications were observed between similar diseases, indicating areas for improvement. Despite these challenges, the results demonstrate the effectiveness of deep learning for fish disease classification.