论文部分内容阅读
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。
Aiming at SAR target identification problem in complex scenes, this paper proposes a SAR target identification algorithm based on Bag-of-Words (BoW) model. In the low-level feature extraction stage of BoW model, the algorithm uses SAR-SIFT feature to describe the shape information of local area. At the same time, a new set of local SAR image features based on traditional feature identification is used to describe the contrast and texture information of local area . For the fusion of multiple underlying features in Bo W model, the algorithm uses the feature fusion of the image layer to generate the global identification feature of the image. The weight coefficients of the BoW model features are trained by the L2 norm-constrained multi-core learning method . Experiments on the target of Mini SAR SAR image data show that SAR target recognition algorithm based on multi-feature fusion Bo W model has better performance compared with the traditional algorithm based on Bo W model and traditional IBD feature. Identify performance.