Random forest and watershed transformations for mapping quasi-circular vegetation patch using fused multispectral CBERS-04 images

Abstract

To identify the quasi-circular vegetation patches (QVPs) and monitor their pattern dynamic is the most essential procedure for studying the establishment, encroachment, and maintenance of the QVPs. High resolution satellite remotely sensed technique is a cost-effective approach to distinguish the QVPs. However, the adhesion between the QVPs or between the QVPs and the vegetations of other shapes makes the detection accuracy of the QVPs low. This study used two-seasonal CBERS-04 pansharpened multispectral images to detect the QVPs by integrating random forest classification and watershed transformation image segmentation technique. The results showed that the method proposed by this study could raise the detection accuracy of QVPs, and 5 m spatial resolution CBERS-04 pansharpened imagery may be enough to obtain the number of the QVPs, but not sufficient to calculate the area and shape of QVPs. In the future, more advanced image segmentation algorithms and finer resolution images should be applied to detect the QVPs.