Cloud based Data Acquisition and Annotation Architecture for Weed Control

Abstract

In this paper we present a short evaluation of a cloud based architecture for data acquisition and annotation. We evaluate the implemented system for annotation and give initial results on the ability of the system to produce accurate labels on the data. The used data is consisted of plant field images. The users partially annotate the data and we use segmentation algorithms for enriching the annotation of the images. We compare three different segmentation algorithms used for the annotation. The results show that Grabcut algorithm is better than Watershed and nearest-neighbor approaches, but there is still room for improvement.