The goal of this research is to create a lightweight deep learning system for crop disease classification using the PlantVillage dataset. MobileNetV2, EfficientNetB0, and ResNet- 18 are a group of effective convolutional neural network models. In order to improve the models’ robustness, they were trained using augmented images, and their accuracy, precision, recall, and F1-score were compared. The models were subsequently optimized using quantization at TensorFlow Lite float16 and INT8. With fewer misclassifications, EfficientNetB0 identified a majority of 38 disease classes. Quantization demonstrated the models’ applicability in a low-power system by greatly lowering their size and providing a viable inference runtime. Additionally, a simple web interface for uploading leaf photos for automatic disease detection has been developed. The trials demonstrated that small-scale models may identify plant diseases effectively and efficiently enough to be used in edge-based or mobile agricultural systems.