利用高斯混合模型的SAR图像目标CFAR检测新方法

SAR（合成孔径雷达）图像杂波分布模型种类繁多且对实际地物的建模能力有限。在使用基于杂波统计模型的CFAR(恒虚警率)算法对SAR图像进行目标检测时，杂波统计模型的失配会导致检测结果产生较大的CFAR损失，算法精度不高。提出了一种基于高斯混合模型的CFAR检测新方法。该方法以理论上可以拟合任意形状概率密度分布的高斯混合模型对实际SAR图像的背景杂波进行拟合，利用拟合后得到的分布模型，根据CFAR检测的原理推导出目标检测阈值的计算公式完成目标的检测。新方法对服从不同分布模型的背景杂波，使用形式上统一的模型进行描述，克服了CFAR检测高度依赖背景杂波分布的缺点，提高了CFAR的通用性。实验结果表明，即使在背景杂波类型未知的情况下，新方法依然得到了良好的目标检测效果。

A Novel CFAR Algorithm for Detecting Targets in SAR Images Using Gaussian Mixture Model

Clutters in SAR (synthetic aperture radar) image is various and complex. So different kinds of distribution models should be used for CFAR (constant false alarm rate) based target detection of airborne SAR images, which increases the difficulties and complexities of automatic target detection. Gaussian mixture model (GMM) is an extension to Gaussian probability density function. Theoretically, it can approximate any distribution smoothly. In this paper, clutter distributions of SAR image are estimated using the GMM. And a novel CFAR threshold expression was proposed. Describing different clutter distributions with one uniform model, the proposed CFAR detector is more universal since it is less dependent on clutter distributions. The experimental results show that, even though the clutter distributions are unknown, the proposed technique still has a nice performance.