On Statistical Approaches To Feature Extraction And Recognition For Target Above The Water

In recent years, with the further research and exploration to oceans, It seems more important to control the information of the ocean in our real life, in which the automatic target recognition(ATR) of surface targets is one of the most important issues.As the surface of the water is more complex and more unpredictable, this has brought considerable difficulties effort to automatic target recognition of surface targets. The two traditional approaches are holistic in nature and are based on moment invariants and principal component analysis (PCA); performance is evaluated under the simulated conditions of imperfect localization by a region of interest (ROI) algorithm, so the two methods of identification are in great limitations.Firstly, the basic theory, methods and applications of image pattern recognition are analyzed in this paper, combined with the requirements of surface target identification, the image pre-processing content is studied, mainly for the impulse noise filtering and image segmentation to explore, and based on experimental results for determining the method of image filtering and segmentation. Secondly, through the research of target recognition method based on moment invariants and principal component analysis, and the analysis of the results, we present a methodical evaluation of the performance of new approaches based on local features (object parts) and is comprised of a block-by-block 2D Hadamard transform (HT) coupled with a Gaussian mixture model (GMM) classifier. Experiments show that the proposed approach has good robustness to clipping and, to a lesser extent robustness to scale changes.The moment invariants based approach achieves poor performance in advantageous conditions and is easily affected by clipping and occlusions. The PCA based approach is highly affected by scale changes and clipping, while being relatively robust to occlusions and noise. Furthermore, we show that the performance of a silhouette recognition system subject to mismatches between training and test angles of silhouettes (caused by an out-of-plane rotation) can be considerably improved by extending the training set using only a few angles which are widely spaced apart. The improvement comes without affecting the performance at"side-on" views.