ShrimpXNet: A Transfer Learning Framework for Shrimp Disease Classification with Augmented Regularization, Adversarial Training, and Explainable AI

AI can improve disease detection in aquaculture: The study achieved 96.88% accuracy in automatically classifying shrimp diseases using deep learning, demonstrating that AI systems can provide "timely and accurate detection" to help farmers identify outbreaks quickly—a critical capability given that diseases "pose a major challenge to sustainable shrimp production."

Robustness testing makes AI systems more reliable: The researchers used adversarial training (FGSM) and advanced data augmentation techniques (CutMix and MixUp) to stress-test their models and prevent overfitting, showing that responsible AI development requires deliberate efforts to ensure systems perform reliably under varied real-world conditions, not just ideal laboratory scenarios.

Explainability helps build trust in AI decisions: The team applied multiple explanation methods (Grad-CAM, Grad-CAM++, XGrad-CAM) to visualize which regions of shrimp images the model focused on when making classifications. This transparency is essential for farmers and veterinarians to understand why the AI made a particular diagnosis and verify that it's looking at relevant features rather than spurious patterns.

Smaller, efficient models can deliver strong performance: ConvNeXt-Tiny outperformed larger models while being more computationally efficient, suggesting that responsible AI doesn't always require massive systems—practical solutions for resource-limited farming communities can be both accurate and accessible.