Identification and Prediction System of Tomato Leaf Diseases Based on YOLO and Random Forest

Tomato crop failure in Indonesia caused by leaf disease can reach 78%, and manual visual monitoring by farmers is often delayed and subjective. This study designs and evaluates a portable, IoT-based diagnostic and prediction system for tomato leaf disease that can operate independently in the field. The system integrates a Raspberry Pi 4 single-board computer with a Raspberry Pi Camera V2, a DHT22 air temperature/humidity sensor, and a capacitive soil-moisture sensor. Two computational models run on-device: a YOLOv5 convolutional neural network that classifies leaf images into healthy, early blight, late blight, and septoria leaf spot; and a Random Forest classifier that predicts disease potential from air temperature, air humidity, and soil moisture. Results are transmitted via a 4G module to a Firebase cloud database and displayed on a real-time dashboard. Functional and field tests validated camera resolution, sensor accuracy, model accuracy, processing latency, and transmission delay. The YOLOv5-small model classified leaf images with 85.4% accuracy, while the Random Forest model predicted disease potential with 97% accuracy. On-device dual-model inference averaged 7.26 seconds and cloud transmission delay averaged 8.8 seconds, both within the 10-second design target. Field validation on a working tomato farm produced an average diagnostic correctness of 85.53%. The dual-model, off-grid-capable IoT system is technically feasible for real-time, field-deployable tomato leaf disease diagnosis and can help farmers act earlier on emerging infections, reducing potential crop loss.