Identification and estimation of the extent of factors influencing maize production using images from Unmanned Aerial Systems

Abstract

Precision agriculture is becoming increasingly popular. From the use of automated irrigation systems to automated pesticide distribution using drones (Unmanned Aerial Systems (UASs)) spanning to yield-estimation through the use of image sensors mounted on tractors. UASs are becoming an increasingly common and convenient way of obtaining information about crops. However, the information gathered from these UASs is still manually interpreted utilising visual inspection of the images. This study investigates the possible automation of identification and estimation, of the extent of damages on food crops using UASs. This study focuses on maize as a food crop, and water stress as primary means of damage. Other types of damage are also investigated, however, they are not included in this experiment. Still images are captured and stitched together using GPS-coordinates and camera parameters. A brief investigation on the topic of image stitching is provided with findings on the particular method(s) used. Following the stitching process, investigation and experimentation on vegetation segmentation are discussed along with the findings of this particular case. Finally, a neural network is experimented with to classify crop areas as 'damaged' or 'healthy'.