Gaussian mixture models for image-based cereal plant canopy analysis

Abstract

In this paper, we report our results of applying Gaussian Mixture Models (GMM) to the analysis of the canopy of cereal plants grown in competitive environments, such as large bins. We will particularly focus on the segmentation problem, i.e. separating the plant regions from the other image regions, such as soil, water pipes, and bin walls. We will show that GMMs, which require few training images, provide a flexible and efficient tool for high throughput segmentation at various growth stages and even in the presence of complex background. We discuss various implementation issues and provide results on a large scale experiment, where cereal plants of different genotypes are grown in large bins and subject to two different treatments (well watered and under drought stress).