Real-time plant image segmentation algorithm under natural outdoor light conditions

Variable rate pesticide application holds Great potential in precision agriculture where application efficiency depends mainly on plant recognition. The segmentation of the plant from the background is key to plant recognition. The work presented in this paper is a real-time segmentation algorithm that was developed to improve the quality of plant segmentation under natural outdoor light conditions. The plant images were greyed with the colour features H, a*, I3 and Cr respectively, and the plant was segmented from the grey images with a threshold that was computed by an iterative method. Experiments showed that segmentation was generally of good quality if the background was bare soil. The quality of segmentation declines if the images are greyed by H and Cr, and the quality remains stable if they are greyed by a* and I3 when the background is complicated, and, especially with heavy shadows. With respect to segmentation speed, that greyed by Cr was the fastest and that greyed by a* was slowest amongst the four.