Study On The Methods Of Ship ROI Extraction In Optical Satellite Remote Sensing Image

Ship detection is one of the most active research topics in the field of remotesensing, and the aim of ship detection is to localize and interpret ship targets in marineenvironment. This paper mainly discusses the technology of ship ROI (Regions ofInterest) extraction in Optical satellite remote sensing image. The main work andcontribution of this paper can be summarized as follows:1. A method of ship ROI extraction from optical satellite remote sensing imagerybased on mathematical morphology and Gabor filter is proposed in this paper. Firstly,the Tophat operator is used to extract the potential Regions of Interest (ROI). Based onthe mechanism of human vision perception, Gabor filters is then applied to reduce thenegative influence of non-target points and enhance the target points. Finally, fractalfeature is used to detect targets by discriminating targets from background.Experimental results based on optical satellite remote sensing images show that ourmethod is effective and feasible.2. Ship target segmentation in optical satellite remote sensing images meets aserious challenge, owing to the variations of light conditions, visual angle and theshadow problem. To address these difficulties, this paper presents a ship segmentationmethod based on Chan-Vese (CV) model and shape information. First, Kernel PrincipleComponent Analysis (KPCA) is used to extract shape information of optical satelliteremote sensing ship targets. The obtained shape energy term and the CV energy termare then combined into a new ship segmentation model. To ensure the numericalstability of symbol function of our model, we also present a new Heaviside function inthe evolvement process of level set. Experimental results show that our method cancontribute better segmenting effect than original CV model.3. To determine whether the target in ROI is a ship or not, this paper investigate thetechnology of target discrimination and implemented a feature-based ship targetdiscrimination method. First, three kinds of features including gray, shape and textureare extracted from ROI. Different schemes of feature combination are then used to traindifferent SVM classifiers. After choosing the best feature combination scheme andSVM classifier, we can obtain an effective ship target discriminator for ship targetdiscrimination.