<![CDATA[ This study aims to develop an early detection system for rice plant diseases using a machine learning (ML) approach based on edge computing with ESP32-S3 Cam devices and the Edge Impulse platform. This system is expected to provide an efficient and cost-effective solution for detecting rice diseases in agricultural areas with limited internet and electricity access. In this study, CNN and MobileNetV2 models were used to classify rice leaf diseases, including brown spot, tungro, and blight, achieving 92.73% accuracy on the test dataset. This system is designed with an offline-first principle, allowing the device to operate locally by optimising power and memory usage. The model, which is optimised through quantisation and transfer learning, is small in size, only about 587 KB, and can be operated on devices with limited resources. In addition, this system can send notifications via Telegram and Google Sheets when connectivity is available. Field test results show that the system performs well across various environmental conditions, including low light and high humidity, with a detection accuracy of 90-95%. With innovations in lightweight ML models and edge computing, this study contributes to improving agricultural efficiency in Indonesia, especially in addressing the challenges posed by climate change that affect rice production. This research also provides insights for the further development of smart farming systems integrated with IoT technology for real-time disease detection.]]>
