According to the FAO of the UN, availability, access, utilization, and stability are the four pillars of food security that largely depend on sufficient, safe, and nutritious food. Detecting plant disease in advance might be a measure of resilience to the future disruption or unavailability of food supply. Due to the notable performance through highly accurate mechanization, deep learning-based methods have been applied to automatically identify and diagnose plant disease that can improve efficiency and productivity. The work prioritizes Transfer Learning of VGG16 for predicting potato blight disease. The model's weights are pretrained on ImageNet, which can be extracted from specific features of small datasets. The implemented approach presents a significant performance improvement on a self-prepared dataset. After completing the necessary training and testing process, 96.88% accuracy was achieved by the model. Experimental results are compared with well-established models, which concludes that the model performs better in classifying potato leaves blight diseases. © 2021 IEEE.