HSL Color Space for Potato Plant Detection in the Field

Abstract

This research paper discusses a vision system and image processing algorithms for an autonomous vehicle to be implemented for precision agriculture purposes. This system is a part of a larger project, to detect and remove potatoes infected by a commonly occurring virus (PVY – potato virus Y). For the detection and removal of infected plants, first, an unmanned aerial vehicle (UAV) equipped with a hyperspectral camera and a high precision GPS, will fly over the potato field collecting images of the plants. Using custom image analysis, the GPS location of the sick plant is identified and sent to an autonomous ground vehicle (AGV). This AGV will then navigate to the target location and rogue the infected plant automatically. The RTK GPS used here has an error of about 10cm. After the AGV reaches the target location the automatic roguing mechanism will still need to identify the sick plant. Potato seeds are planted at an average distance of about 30 centimeters, but in reality, this distance may vary significantly in the field. To positively identify the sick plant in real-time, a special image processing system was designed to detect and position the rouging arm over the center of the sick plant. This system uses an 8 Megapixel Pi camera to find the center of the target plant looking down. This system needs to work with high accuracy in a potato field where changing sunlight and weather conditions would hamper proper identification, HSL (hue, saturation, and lightness) format of images was used for better color detection. Two methods for finding the center of the plant were compared. These were compared to positive detection rates for various light levels, a variety of leaf colors, and expected location as opposed to actual plant location.