Detection of Small Portions of Water in VIS-NIR Images Acquired by UAVs

摘要

Water bodies detection in aerial images is a problem widely known and explored in the literature. However, the detection of small portions of water in satellite images is not usual due to their low spatial resolution. With the increasing use of Unmanned Aerial Vehicle (UAVs), this task becomes feasible from the point of view of spatial resolution, but suffers from the problem of low spectral resolution of the acquired images. In this sense, we have new challenges in dealing with an old problem. In this work we proposed an approach that combines Visible (VIS) and Near Infrared (NIR) aerial images, commonly acquired by sensors used in UAVs, for providing an indicative index for detecting small portions of water. This approach also includes a scheme to reconstitute visible spectral bands contaminated by the use of a special lens, coupled to visible RGB camera, to provide NIR spectral band. Experimental results evidenced the good accuracy of proposed approach in mapping small portions of water such as pools and fountains in VIS-NIR images, even in the cases where there is only a thin layer of water.