Smart aquaculture: Fine-tuned ResNet50 for precision fish disease detection and sustainable health management
Bhuria, R. (2025). Smart aquaculture: Fine-tuned ResNet50 for precision fish disease detection and sustainable health management, in: 2025 IEEE International Conference on Interdisciplinary Approaches in Technology and Management for Social Innovation (IATMSI), 6-8 March 2025, Gwalior, India. pp. 1-6. https://dx.doi.org/10.1109/iatmsi64286.2025.10985132
In: (2025). 2025 IEEE International Conference on Interdisciplinary Approaches in Technology and Management for Social Innovation (IATMSI), 6-8 March 2025, Gwalior, India. IEEE: [s.l.]. ISBN 979-8-3315-2169-1. https://dx.doi.org/10.1109/IATMSI64286.2025, more
|
| Keywords |
Aquaculture
Control > Disease control
|
| Author keywords |
Food Security, Sustainable Agriculture, Food Production, Livelihoods, Nutritional Needs, Environmental Conservation
|
| Abstract |
Although aquaculture is an essential industry for world food security, several fish illnesses that may cause major economic losses and lower fish populations sometimes compromise it. Effective management and mitigating of these diseases depend on timely and correct diagnosis. This article provides a deep learning-based method employing a fine-tuned ResNet50 convolutional neural network (CNN) to classify fish illnesses.The 1,750 high-resolution photographs in the dataset are evenly split among seven classes: Aeromonas, Bacterial Gill Disease, Bacterial Red Disease, Saprolegniasis, Healthy Fish, Parasitic Diseases, and White Tail Disease.Each class comprises of 250 photos. Part of the approach includes complete data preparation comprising image reduction to 224x224 pixels, normalisation, and augmentation techniques like rotation and flips to increase dataset diversity. Comprising seventy percent training, fifteen percent validation, and fifteen percent testing subsets, the dataset gives balanced representation and strong model evaluation.Originally trained on ImageNet, the ResNet50 model is improved utilizing additional global average pooling, dropout, and thick layers especially for the seven-class classification challenge. Particularly in discriminating healthy fish from various bacterial illnesses, the trained model performs remarkably with overall accuracy of 97%, precision ranging from 93% to 100%, recall from 93% to 100%, and F1-scores routinely high across all classes. The confusion matrix reveals the great discriminative capacity of the model by demonstrating few misclassifications, mostly across classes with similar visual symptoms. This work presents the efficiency of fine-tuned ResNet50 in precisely classifying fish diseases, so providing a dependable diagnostic tool that can help aquaculture management in disease control and prevention, so promoting sustainable and lucrative fish farming methods. |
|
