DEVELOPMENT OF MACHINE LEARNING BASED REALTIME IMAGE PROCESSING FOR HONEY SWEET POTATO (IPOMOEA BATATAS) LEAF DISEASE DETECTION SYSTEM

• Main Author: Franciscus Leander C., Jose
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84062
• Institution: Institut Teknologi Bandung

Indonesia’s population is increasing over time. Over the past 10 years there has been an increase of 25 million people. Therefore, the need for food also increase. However, the number of farmer is decreasing over time and coupled with the aging farmer phe- nomenon, a touch of technology is needed which can increase the efficiency of food production. Technologies such as IoT, robots, sensors, and machine learning can be applied in agriculture. This application is called smart agriculture. There are 5 scopes that can be developed in smart agriculture such as precision agriculture, irrigation ma- nagement, soil management, pest and disease management, and crop monitoring. This study focuses on creating real-time image processing which is included in disease dete- ction system for Honey Sweet Potato leaves. This system, hopefully, can detect disease precisely, accurately, and real-time. There are 2 parts developed, namely disease dete- ction system and data processing system. Within the application, the disease detection system captures crop images, predicts the disease, and sends it to the server. Then, the server stores and processes the data to be shown to the user in the form of disease spre- ad graph and heatmap for all detected diseases. To make a prediction, disease detection system takes 1.16 seconds in average to process with the error from 20% for necrosis to 121% for gulma. From data processing system, the most dominant disease detected is leaf spot which is 51.4% of all disease detected. Heatmap shows that Chlorosis and Leaf Spot spreads in almost all parts of the crop. While Gulma, Necrosis, and Pest only occur in few location of the crop.