Globally, shrimp aquaculture is a vital source of food, but disease outbreaks present serious economic challenges. Traditional diagnostic techniques, such as visual inspection and polymerase chain reaction (PCR), are limited in their ability to detect diseases in real-time because they are either resource-intensive or prone to human error. In order to overcome these obstacles, we propose FeatherNetX, a lightweight deep learning framework, designed for automated shrimp diseases classification and deployment in resource-constrained settings. Black Gill (BG), White Spot Syndrome Virus (WSSV), Yellow Head Virus, and Healthy classes were among the publicly accessible shrimp disease image datasets used to train the model using 5-fold cross-validation approach. FeatherNetX outperforms other models with an average accuracy of 93% ± 0.059, a small model size (0.739 M parameters, 2.82 MB) and low computational cost (0.48 GFLOPs) while outperforming traditional architectures in terms of efficiency. In order to improve interpretability, disease-relevant regions were visualized using Grad-CAM++, which demonstrated a high degree of correspondence between activated regions and ground-truth disease areas. Additionally, the model was incorporated into a desktop application that could classify images in real-time and offline into specific classes with confidence reporting. It achieved 94% accuracy on test images that were unseen and took an average of less than 0.2 seconds per image to classify. This study bridges the gap between deep learning research and practical aquaculture practice by offering a strong framework for automated classification of shrimp disease through the combination of lightweight architecture, model interpretability, and practical deployment.
Sharma et al. (2026) studied this question.