Classification algorithm for embedded systems using high-resolution multispectral data

The extraction of remote sensing signatures from a particular geographical region allows the generation of electronic signature maps, which are the basis to create a high-resolution collection atlas processed in discrete time. This can be achieved using an image classification approach based on pixel statistics for the class description, referred to as the multispectral pixel neighborhood method. This paper explores the effectiveness of this approach developed for supervised segmentation and classification of high-resolution remote sensing imagery using SPOT-5 data. Moreover, an analysis of the proposition for implementation as an embedded system is provided, to improve the processing time and reducing computational load, using a scheme based on hardware/software codesign techniques. Simulations are reported to probe the efficiency of the proposed technique.