基于多特征融合的目标区域快速筛选与场景鉴别方法

Method for Rapid Target Region Screening and Scene Discrimination Based on Multi-Feature Fusion

摘要:

针对合成孔径雷达（synthetic aperture radar，SAR）图像目标识别面临的虚警率高、场景复杂等挑战，提出了一种基于多特征融合的目标区域快速筛选与场景鉴别方法。该方法采用子图分割、像素级聚类和灰度阈值判别相结合的策略，实现了基于匀质杂波剔除的目标区域快速筛选。同时，该方法利用大型建筑在SAR图像中表现出的强散射点密集、阴影显著、几何形状规则等特征，实现了基于强散射点分析的大型建筑精确鉴别。实验结果表明，所提方法能够有效剔除由人工及自然纹斑噪声、杂波造成的虚警场景，显著降低目标识别虚警率，在宽幅扫描成像模式下实现了目标区域的快速搜索和筛选，为后续目标识别提供了高质量的数据输入。

Abstract:

To address the challenges of high false alarm rate and complex scenes in target recognition for synthetic aperture radar (SAR) images, a method for rapid target region screening and scene discrimination based on multi-feature fusion was proposed. The method adopted a combined strategy of sub-image segmentation, pixel-level clustering, and gray-level threshold discrimination，to achieve rapid screening of target regions based on homogeneous clutter elimination. Meanwhile，the method utilized the characteristics of large buildings in SAR images, such as dense strong scattering points, significant shadows, and regular geometric shapes, to achieve accurate identification of large buildings based on strong scattering point analysis. Experimental results demonstrate that the proposed method can effectively eliminate false alarm scenes caused by artificial and natural speckle noise and clutter, significantly reduce the false alarm rate of target recognition, and achieve rapid search and screening of target regions under wide-area scanning imaging mode, thereby providing high-quality input data for subsequent target recognition.