Using clustering algorithms to segment UAV-based RGB images

This article describes the implementation of two segmentation algorithms in combination with the Triangular Greenness Index (TGI) derived from images obtained from an unmanned aerial vehicle (UAV), with the objective of segmenting shadow, soil and vegetation data obtained from a commercial vineyard cv. Cabernet Sauvignon. The importance of this segmentation lies in the recent development in tools that allow remote monitoring of crops but that nevertheless still have unresolved methodological aspects. The precise differentiation of these classes would allow the development of more complex monitoring techniques based on multispectral and thermal sensors. The results of this investigation showed that both k means and Clustering Large Applications (CLARA) allowed to differentiate three classes in the images corresponding to soil, shade and vegetation. However, CLARA showed a better performance when differentiating the layer corresponding to vegetation.