Machine learning techniques for the assessment of citrus plant health using UAV-based digital images

Abstract

This paper presents the use of machine learning techniques for the development of a methodology for the analysis of digital images of citrus plants collected from unmanned aerial vehicles (UAVs). Proven ground based sensors including a chlorophyll meter, water potential meter, and spectroradiometer are used to evaluate the condition of the plants, thus providing the ground truth. The collected images and ground truth data are then used as training data to the machine learning models, which are validated using a separate set of data. For our models, we evaluate several machine learning techniques from simple linear regression to convolutional neural networks. The overall goal is to develop a solution for monitoring plant health that can readily and cost-effectively be used by farmers to determine nitrogen and water stresses in plants. Such a system will aid in the conservation of physical resources while reducing human labor and the environmental impact of chemicals.