Publication

Abstract : Monitoring of fish health through precise disease detection is essential in sustainable aquaculture management. In this article, a deep learning method is proposed to detect and classify fish diseases from image data of seven categories, consisting of six disease classes and one for healthy fish. Several Convolutional Neural Network (CNN) architectures are compared, including ResNet (18, 34, 50), MobileNet (V2, V3), and Inception (V3, V4), considering the classification accuracy and interpretability. To improve model explainability, Gradient-weighted Class Activation Mapping (Grad-CAM) is used to highlight the salient areas that drive CNN predictions. Experimental results indicate that MobileNetV3 and InceptionV3 obtain better test accuracies of more than 98%, while providing accurate localization of affected areas, allowing for more explainable and reliable predictions. The method is capable of continuous automatic fish health monitoring, thus diminishing manual inspection needs and potentially increasing the efficacy of aquaculture management.