Multitarget geolocation via an agricultural octorotor based on orthographic projection and data association

Abstract

In recent years, unmanned aerial vehicles with onboard spectral sensors have been used in detecting diseases in the agricultural fields. Geolocation, i.e. calculating the global coordinate of identified diseased regions based on images taken, is an important step in automating such a scouting operation. In this paper, the problem of geolocating multiple diseased regions in an image is studied. Based on the assumptions of stationary, two-dimensional shallow target plants and hover flight, an orthographic projection-based measurement model is developed. A probabilistic data association method is used to analyze the measurements from different target sources and a Kalman filter is designed to estimate the suspected diseased leaves’ position. To the best of the authors’ knowledge, it is the first time a data association technique is used in for locating multiple-diseased plants in agriculture applications. Additionally, the designed Kalman filter is based on conditions pertinent to small crops and is less computationally intensive than the typically used extended Kalman filter. Both simulation and ad hoc experiments are used to validate the proposed multitarget geolocation algorithm.