一种基于稀疏化核方法的红外强杂波背景抑制算法

An IR Strong Clutter Background Suppression Algorithm Based on Sparse Kernel Method

杂波背景抑制一直是红外弱小目标检测面临的难题.背景抑制可分为背景预测和差分滤波两步.针对强杂波背景呈现非线性分布的特征,提出了一种基于稀疏化核递推最小二乘(KRLS)算法的非线性背景抑制算法.算法采用监督学习模型,使用序列图像作为训练样本.通过稀疏化控制学习函数的复杂度并剔除冗余信息,不但可以提高学习机器的推广能力,还可以降低运算量.使用真实红外图像对算法进行了测试,并分析了算法参数.实验结果表明:算法可自适应预测不同类型的强杂波背景,并有效抑制背景杂波.

Clutter background suppression is always a difficulty of infrared （LR） dim and point target detection. Background suppression is divided into background estimation and difference filtering. Aiming at the nonlinear distribution of strong clutter background, a spares kemel recursive least squares （KRLS） based nonlinear background suppression algorithm is proposed. This method uses sequence images as training sample in supervised learning model. The complexity of learned function is controlled, and the redundant information is discarded by sparsification. In this way, the generalization of learning machine can be enhanced; moreover, the computational burden can be reduced. In the experiments,real IR images are used to test the algorithm, and the parameters are analyzed. Experimental results show that different kinds of strong clutter background can be estimated, and then be suppressed.