Plant diseases are a major cause of agricultural productivity, burdening it with economic damage. Therefore, accurate early identification is critical for efficient crop management. In this study, we proposed a new deep learning approach for real-time plant disease classification based on the EfficientNetB3 model. This is an implementation of transfer learning using the EfficientNetB3 transfer learning model, fine-tuning using the PlantVillage dataset, which contains images of healthy and diseased leaves of crops such as pepper, potato, and tomato. It achieved a highest accuracy of 96.05%, demonstrating its ability to effectively classify images of diseased plants from healthy plants. Model robustness was ensured through comprehensive evaluation techniques (accuracy assessment, confusion matrix analysis, and loss visualization). The results demonstrate that the EfficientNetB3-based approach surpasses EfficientNet baseline architectures, achieving higher performance, comparable accuracy, and higher efficiency. This work helps create automatic diagnosis systems that can be deployed in edge devices or cloud solutions to monitor the disease in real time. The system assists in decision making regarding when to apply disease manage- ment, optimize crop yield, and minimize losses. Future work will involve optimizing real-time deployment and extension to multicrop disease classification using enhanced deep learning architectures.