Fish diseases pose a significant threat to the aquaculture industry, leading to substantial economic losses and reduced productivity. Current detection methods, relying heavily on visual inspection, are time-consuming and require expert knowledge. This paper proposes an automated fish disease detection system using deep learning, specifically the ResNet50 architecture. Utilizing a dataset of 1750 high-quality fish skin images spanning seven disease categories, the model achieved 96.7% accuracy and an F1 score of 96.7%. This system offers a scalable, efficient, and highly accurate tool for monitoring fish health, reducing reliance on manual diagnosis. Future work aims to integrate IoT frameworks for real-time surveillance and expand dataset diversity to enhance model generalization.