SmartAgriDoc: Offline Plant Disease Detection with Improved Accuracy

This paper investigates the use of machine learning (ML) methods for offline plant disease detection using leaf images captured through a mobile device.Data was obtained from the PlantVillage dataset, along with fieldcollected images representing real agricultural environments.The objective of this research is to develop a light-weight model that can accurately classify plant diseases without network connectivity, enabling realtime diagnosis for farmers in rural regions.Multiple CNN-based architectures and Mobile-AI models were compared.MobileNetV2, EfficientNet-Lite, and a custom CNN demonstrated strong results, with MobileNetV2 showing the most efficient trade of between accuracy and inference speed on mobile hardware.While the dataset included varying lighting conditions andheterogeneous crop species, the optimized SmartAgriDoc model performed reliably in both controlled and field scenarios.These results support the development of AIpowered.offline disease identification tools for practical agricultural use.