A multi/hyper-spectral imaging system for land use/land cover using unmanned aerial systems

During last years the automated classification enabled by Unmanned Aerial System (UASs) captured the interest of many investors and researchers. Several applications could be explored ranging from precision agriculture to Land Use / Land Cover (LU/LC) thematic mapping. The use of UASs is novel when we consider the very high resolution (VHR) multi-hyper spectral sensing of a given area. Remote sensing and LU/LC have a strong intersection from many years even if only in the last period VHR images over several bands are applicable considering the technological progress. Today it is possible to acquire data in Visible (VIS) - Near InfraRed (NIR)-Short Wave InfraRed (SWIR) by using compact and low cost sensors that could be integrated into small size UAS. These sensors overcame the main limitations of classical remote sensed data from satellite increasing the spectral, spatial and temporal resolution also reducing the influence of clouds and water vapor on atmospheric absorption. In particular, in this paper we propose an imaging system to perform analysis from thematics maps derived from hyper-spectral radiometers and multi-spectral cameras mounted on UAS. The high spectral and geometric resolution enhance the level of details of a LU/LC maps. We propose also a novel method to fast classify data by using an improved version of the k-means algorithm. The proposed method significantly reduces the computational time especially for very large high-resolution data-set. Comparison of k-means over regular grid and quadtree decomposition are discussed.