基于Gabor滤波器和局部纹理特征提取的SAR目标识别算法  

A Method of SAR Target Recognition Based on Gabor Filter and Local Texture Feature Extraction

作　　者：王璐[1] 张帆[1] 李伟[1] 谢晓明[1] 胡伟[1] 

机构地区：[1]北京化工大学信息科学与技术学院,北京100029

出　　处：《雷达学报（中英文）》2015年第6期658-665,共8页Journal of Radars

基　　金：国家自然科学基金(61302164);中央高校基本科研业务费专项资金(YS1404);北京高等学校青年英才计划(YETP0500)~~

摘　　要：该文提出了一种基于Gabor滤波器和Three-Patch Local Binary Patterns(TPLBP)局部纹理特征提取的合成孔径雷达(Synthetic Aperture Rader,SAR)图像目标识别算法。首先,利用Gabor滤波器对SAR图像在不同方向上进行滤波,增强SAR图像中目标及其阴影的关键特征;然后,利用TPLBP算法对Gabor滤波之后的图像进行局部纹理特征提取,该算法克服了Local Binary Patterns(LBP)算法无法描述大范围领域纹理特征的缺陷,并且保持了LBP旋转不变的特性,减少了SAR图像目标方位变化对识别效果的影响;最后利用极限学习机(Extreme Learning Machine,ELM)分类器实现目标识别。该文通过MSTAR数据库中的3类SAR目标识别实验验证了该算法的有效性。This paper presents a novel texture feature extraction method based on a Gabor filter and ThreePatch Local Binary Patterns（TPLBP） for Synthetic Aperture Rader（SAR） target recognition. First, SAR images are processed by a Gabor filter in different directions to enhance the significant features of the targets and their shadows. Then, the effective local texture features based on the Gabor filtered images are extracted by TPLBP. This not only overcomes the shortcoming of Local Binary Patterns（LBP）, which cannot describe texture features for large scale neighborhoods, but also maintains the rotation invariant characteristic which alleviates the impact of the direction variations of SAR targets on recognition performance. Finally, we use an Extreme Learning Machine（ELM） classifier and extract the texture features. The experimental results of MSTAR database demonstrate the effectiveness of the proposed method.

关 键 词：SAR目标识别 GABOR滤波器 Three-Patch LBP 特征提取 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]