Deep learning-driven IoT solution for smart tomato farming.

Abstract

The rising food demand and challenges with respect to the climate have made precision agriculture (PA) vital for sustainable crop production. This study presents an IoT-based smart greenhouse platform tailored for tomato farming, integrating environmental sensing and deep learning. The system employs ESP32-based wireless sensors to collect real-time data on soil moisture, temperature, and humidity; this data is transmitted to a cloud dashboard (ThingsBoard) for remote monitoring. A Raspberry Pi equipped with a Pi Camera and a YOLOv8 model classifies tomato ripeness stages-green, half-ripened, and fully ripened-using real greenhouse images. Model optimizations, including quantization, pruning, and TensorRT, improved inference speed by 35% while maintaining 52.8% classification accuracy during our initial stage of the project. Energy profiling revealed daily consumption of 8.91 Wh for the ESP32 sensors and 78 Wh for the Raspberry Pi. This prototype demonstrates real-time monitoring, high model precision, and practical energy insights, paving the way for multi-node scalability and edge AI enhancements. Future work will explore incorporating Edge TPU for faster on-device processing, LoRa for low-power, long-distance data transfer, and automated control of irrigation and ventilation systems to realize a fully autonomous smart greenhouse.

Plant phenotyping relevance

トマト果実の成熟段階という植物状態をカメラ画像とYOLOv8で推定し、モデル最適化・精度・推論速度を評価しているため、画像ベースの表現型取得が中心的です。

Code and data availability

The paper describes authors' own real greenhouse images and a YOLOv8 ripeness classifier, but no public deposit of these images, sensor data, code, or trained models is stated. The only dataset URL (Laboro-Tomato on Kaggle) is a cited third-party public dataset (reference 7), not a paper-specific asset, and no code or

No evidence-backed public reproduction asset is currently recorded.