基于多特征融合词包模型的SAR目标鉴别算法

SAR Target Discrimination Algorithm Based on Bag-of-words Model with Multi-feature Fusion

针对复杂场景中的SAR目标鉴别问题,该文提出一种基于多特征融合词包(Bag-of-Words,Bo W)模型的SAR目标鉴别算法。在Bo W模型底层特征提取阶段,算法采用SAR-SIFT特征描述局部区域的形状信息;同时,采用该文基于传统鉴别特征提出的一组新的SAR图像局部特征描述局部区域的对比度信息和纹理信息。对于Bo W模型中多个底层特征的融合,算法采用图像层的特征融合方式生成图像的全局鉴别特征,其中各单底层特征Bo W模型特征的权系数通过L2范数约束的多核学习方法训练得到。在Mini SAR实测SAR图像数据上的目标鉴别实验表明,与基于传统鉴别特征以及单底层特征Bo W模型特征的鉴别算法相比较,该文基于多特征融合Bo W模型SAR目标鉴别算法具有更好的鉴别性能。

In order to solve the SAR target discrimination problem in the real complex scenes, a SAR target discrimination method is proposed based on Bag-of-Words（Bo W） model with multiple low-level features fusion. In the low-level feature extraction stage of Bo W model, the SAR-SIFT feature is utilized to describe the shape information of local regions of an image sample. And also, a set of new local descriptors is used to capture the contrast information and the texture information of the local regions, which is extracted based on the traditional target discrimination features. For the fusion of different low-level features in Bo W model, the image-level feature fusion strategy is implemented to generate the image global feature, which is realized by the Multiple Kernel Learning（MKL） method with L2-norm regularization. Experimental results with the Mini SAR real SAR dataset show that the proposed SAR target discrimination algorithm based on Bo W model with multi-feature fusion achieves better discrimination performance compared with methods based on the traditional discrimination features and the Bo W model features using single low-level descriptor.