The Ship Detection Based On Complex Background Optical Remote-sensing Images

With the development of satellite remote sensing technology, large amount ofhigh-resolution remote sensing images can be acquired. Using them, automatic, dynamic andfast ship detection in a wide range of remote sensing images has a great application space inboth the military and civilian fields. The major research of this thesis is based on opticalremote sensing images under the complex background automatically ship detection.Considering the practical application and complex background conditions, in order to improvethe adaptability and practicability of the system, two key technologies are systematicallystudied including candidate region detection and candidate region discrimination, alsointroducing visual attention mechanism used in ship detection algorithm.This paper first introduces the application background, purpose and significance ofoptical remote sensing images ship detection technology, and then summarizes the currentresearch status both at home and abroad, and then designs the overall ship detection algorithmframework. In ship candidate region detection stage, to the optical remote sensing imagespreprocessing, by using median filter for image denoising and OTSU algorithm forsegmenting land and sea. Drawing on classical CFAR algorithm, introduced the doubleparameter CFAR algorithm to extract the target candidate region. Reference to thecharacteristics of the human eye visual attention mechanism, introduced saliency map fastgeneration method based on adaptive morphology. In combination with clutter rate coefficientfrom adaptive control structure element size of Top-Hat operator, and then through a set ofgabor filters simulation human visual perception characteristics, achieve ship detection whileeffectively inhibit complex background. In ship candidate region discrimination stage,summarizes the existing discrimination techniques, using a decision templates basedensemble technology for candidate region discrimination.With the support of a large number of optical remote sensing images data, through thesimulation experiments of ship detection, adjust algorithm combined with the actual to ensurethe effectiveness and adaptability.