Pomegranate farming is greatly confronted by foliar diseases that largely affect yield, fruit quality, and economic outcomes. Traditional approaches to disease identification based on manual inspection by farmers are not just time-consuming and labor-intensive but also subjective and prone to inaccuracies, especially when extrapolated to large farms. In response to these drawbacks, in this research an intelligent vision system is proposed utilizing advanced deep learning algorithms for autonomous real-time identification of pomegranate leaf diseases. The proposed approach presents excellent performance over varying field conditions with high reliability, coping with variability in light conditions, leaf orientation, and stages of disease advancement. Comprehensive testing was performed with field-acquired datasets spanning various types of diseases and normal samples, and the system attained higher classification accuracy than other methods. The major contribution of this work is in its innovative combination of lightweight structure design with adaptive learning to achieve an end-to-end solution that outperforms traditional methods in terms of operational efficiency, scalability, and practical usability with minimal use of computational resources.