Vegetation segmentation robust to illumination variations based on clustering and morphology modelling

Vegetation segmentation from images is an essential issue in the application of computer vision in agriculture. In this paper, we present a new vegetation segmentation method based on Particle Swarm Optimisation (PSO) clustering and morphology modelling in CIE L*a*b* colour space. At the off-line learning stage, a new method is put forward to determine the clustering number. Secondly, the tools of morphological dilation and erosion are employed to establish the vegetation colour model. At the online segmentation stage, the PSO-based k-means is used to cluster the vegetation image into vegetation classes and non-vegetation classes. Afterwards, the established colour model is used to distinguish the vegetation classes and give the segmentation result. In the experiments, the proposed method was applied to segment 200 smaller regions of the full camera images of rice and 100 smaller regions of the full camera images of cotton. The means of segmentation qualities reached 88.1% and 91.7% respectively. Moreover, the proposed method was compared with three well-known vegetation segmentation methods and two skin segmentation methods. Experiments demonstrate that the proposed method yielded the highest mean of segmentation qualities and lowest standard deviations of segmentation qualities. In addition, the vegetation colour models built with different structuring element types are analysed. (C) 2014 IAgrE. Published by Elsevier Ltd. All rights reserved.