Feature Extraction And Recognition Of Ships Using ISAR Images

ISAR can acquire image of long-distance moving target such as ships,aircraft et al,also could get the target image from different angle of view. As of it can detect and track different target,it is widely used in both military and civilian domains. This paper carries research and study about ship features and recognition using the characteristics of ISAR.ISAR technology has a well development by now. This method is used in the research to get ship target. Three kinds of ship model have been used as recognition of the object and establish all types of dot matrix model. Give the simulation of ships based on ISAR image Using Rang-Doppler algorithm. Due to the ship structure and complicated movement and the characteristics of ISAR,the ISAR images are comprised of image from different angle and rotation direction,and each rotation angle of view image has more difference. In order to ensure the generality and availability,a lot of data sample have be used for experiments.Feature extraction is an important step in the recognition. This paper selects three ship features; The General feature,LBP feature and Contour Feature of ship have been used in the research. These features are supplement and promote each other. General feature describes the general situation of the ship target; LBP feature reflect the internal texture characteristics of ship; Contour feature is extracted based on the characteristics of ship image using a fitting method. Then,this paper using the three kinds of features established the features database of ship,respectively.AdaBoost algorithm is a new technology. It is simple and widely used,also has a strong ability of classification. In this paper,three classic AdaBoost algorithms,RAB (Real AdaBoost),GAB (Gentle AdaBoost) and MAB (Modest AdaBoost) have been used. RAB,GAB and MAB used to recognize feature database,respectively. Then,the features are combining to increase the recognition rate. This paper analyses the performance of the algorithm according to the results. Finally,this paper puts forward the step-by-step recognition algorithm in order to enhance the compatibility between classifier and features. This method further enhances the recognition accuracy,and checking its validity. At last,the purpose of this research has been achieved.