Autonomous System for Locating the Maize Plant Infected by Fall Armyworm

Abstract

Digital agriculture helps farmers collect, analyze, and monitor farms while in remote locations, thus spending less time and money while getting more yield. Digital agriculture employs a variety of techniques, including image processing and machine learning. Previous studies have proposed various systems that use unmanned aerial vehicles (UAV) and machine learning techniques to precisely predict maize plants infested by various diseases. In this paper, we propose an extension of the system to locate the exact position of infected maize plants, as well as estimate the size of the infected area. This enables farmers to take appropriate action based on the precise location of the infection, rather than the entire farm. The proposed system performs three tasks: first, it crops UAV images into smaller images and transplants the Global Positioning System coordinates (GPSc) from UAV images into cropped images; second, it extracts the coordinates of the infested maize plant and counts similar GPSc that determine the size of the infected area on every UAV image, and finally, it sends a report to the farmer indicating the infested plants and the size covered by infection on every UAV image. The report helps farmers act quickly and only spray pesticides on infected areas, which saves them time and money. Using a dataset from a maize farm infected by fall armyworm, we show that the system is effective in locating the infected plants and areas.