IoT and Machine Learning System for Early/Late Blight Disease Severity Level Identification on Tomato Plants

Tomato consumption in Indonesia continues to increase every year. Early blight and late blight diseases often attack tomato plants and cause large losses. In this article, an accurate plant disease detection system is designed for the research process of developing high yielding varieties of tomato that are resistant to diseases. The system consists of five subsystems, namely Control, Data Acquisition, Data Storage, Machine Learning, and Data Visualization. Control and Data Visualization are implemented using an Android application. Data Acquisition is implemented with a robotic framework consisting of a sliding cart, an arm, and a camera. The actuators used are stepper motors and servo motors. The data collection is carried out with an Arducam OV5647 with a capturing speed of 8.23 seconds. Data Storage is implemented on three servers: Firebase, CloudMQTT, and Dataplicity, with MQTT and HTTP as the IoT communication protocol. Machine Learning is implemented with an SSD MobileNet V2 FPNLite 640x640 which has an mAP value of 77.25% with an average inference time of 3.71 seconds.