结合BM3D去噪与极限学习机的SAR目标分类方法

An SAR Target Classification Method Based on BM3D Denoising and Extreme Learning Machine

合成孔径雷达(SAR)目标分类一般通过特征提取和分类决策具体实施。采用3维块匹配滤波(BM3D)去噪算法对SAR图像进行处理,减轻噪声干扰的影响。在此基础上,采用极限学习机(ELM)对去噪后的图像进行决策分类。ELM具有很高的分类效率和分类精度,其对噪声的敏感性可通过BEMD去噪算法克服。因此,通过结合BM3D以及ELM的优势可提高目标分类的整体性能。基于MSTAR数据集对提出方法进行测试,结果表明了所提方法的有效性和稳健性。

Synthetic Aperture Radar(SAR)target classification is generally performed by feature extracting and classification decision-making.The Block-Matching and 3D filtering(BM3D)denoising algorithm is applied to SAR images to relieve noise corruption.Afterwardsthe Extreme Learning Machine(ELM)is used to classify the denoised SAR images.ELM has high classification efficiency and precision.In additionits sensitivity to noise corruption can be effectively relieved by Bi-dimensional Empirical Mode Decomposition(BEMD)denoising algorithm.Thereforethe overall classification performance can be enhanced by combining the strengths of BM3D with that of ELM.The proposed method is tested on the MSTAR dataset and the results have proved its validity and robustness.