Maritime Target Recognition Technology Research

The technology of target recognition is one of the most important problems of the research on the area of the modern military information. It has important values when used in collection or monitor of military information, weapon guide. The research of the technology of target-on-sea recognition based on the aerial photos is good help to the defending of the sea power, the finding of the enemy's invading from the sea. But China Army's means of target-on-sea recognition based on the aerial photos is still backward now. This paper study the technology of target-on-sea automatic recognition based on the aerial photos.After studing the nowadays research of the target recognition technology, this paper analyses the characteristics of the target-on-sea and its gray images, in order to extract proper features which are easy to classify the targets. Using these extractive features, a target has been recognized.Firstly, according to the characters of the target-on-sea images in the aerial photos, this paper designs an iterative threshold segmentation approach based on area to obtain the objects of interest. There are two ways, one is that the segmented objects of interest are binary images, one is that the segmented objects of interest are gray images.Secondly, this paper emphasizes on the features extraction from the objects of the images. These discriminative features are invariant corresponding to the objects'rotation, scaling and translation. Considering the moment invariants based on the characteristics of an object's shapes are efficient to target recognition, the moment invariants are presented. For the classification of seeming similar target-on-sea objects, a set of Hu's moment invariants, Zernike's moment invariants and wavelet moment invariants are applied. In fact, most of these moment invariants are extracted from binary images. The images of the target-on-sea have simple background, but the construction of the interesting objects among the images are complex, and different objects have similar shapes, so the moment invariants extracted from binary images are difficult to classify this kind of objects. So this paper presents three kinds of moment invariants when the object images'gray variance is linear. And studies the feasibility and efficiency of the three moment invariants applied to the target-on-sea recognition based on gray images.To get a locally optimum feature set, a feature selection method based on DB...