2DPCA-based Generalized Locality Preserving Projection for SAR Automatic Target Recognition

In this paper,we propose Two-Dimensional Principal Component Analysis (2DPCA)-based Generalized Locality Preserving Projection (2DPCA-based GLPP) method for sAR automatic target recognition. The 2DPCA-based GLPP method projects an input SAR image into the family of projected vectors via horizontal 2DPCA,then project from this space into the classification space via vertical GLPP.2DPCA-based technique is efficient for image representation and aims to preserve the global structure of the original image;GLPP-based technique seeks to preserve the intrinsic geometry of the original image and local structure.So the 2DPCA-based GLPP method not only can preserve the global structure and efficiently represent the original image with lower dimensions,but also can preserve the detailed information of the original image,which is helpful for recognition.Experimental results on MSTAR dataset adopting Nearest Neighbour Classifier (NNC) show that 2DPCA-based GLPP has more discriminating power than PCA,2DPCA,2DLDA and more robustness to the variation of target azimuth. The top correct recognition rate without target azimuth information arrives at 98.61%.
Two-Dimensional Principal Component Analysis (2DPCA) Automatic Target Recognition (ATR) Nearest Neighbour Classifier (NNC)
Wang Tao Huang Yulin Yang Jianyu
School of Electronic and Engineering,University of Electronic Science and Technology of China(UESTC) Chengdu,Sichuan,610054,P.R.China
国际会议
2009 2nd Asian-Pacific Conference on Synthetic Aperture Radar(第二届亚太合成孔径雷达会议)
西安
英文
468-472
2009-10-26(万方平台首次上网日期,不代表论文的发表时间)