论文部分内容阅读
对于背景呈非线性变化的复杂图像,用背景预测的方法对红外点目标进行检测时,传统的线性最小二乘法(Least Squares,LS)的效果比较差。文章使用核方法(Kernel Methods,KMs)推导了最小二乘法的非线性版本:核最小二乘算法(Kernel Least Squares,KLS);进一步推导出了更适合动态系统时序预测的指数加权形式的核最小二乘算法(Kernel Exponential Weighted Least Squares,KEWLS)。提出了一种基于核方法的红外点目标检测算法,先用KEWLS非线性回归算法预测红外图像背景,再通过自适应门限检测残差图像中的目标。非线性函数回归和红外序列图像检测实验表明核方法较大地改进了算法的非线性函数估计与红外背景预测能力。
For complex images with non-linear background changes, the traditional linear least-squares (LS) method is less effective in detecting infrared point targets with background prediction. In this paper, Kernel Least Squares (KLS) is derived by using Kernel Methods (KMs). The kernel of exponentially weighted form that is more suitable for dynamic system timing prediction is further deduced. Kernel Exponential Weighted Least Squares (KEWLS). An infrared point target detection algorithm based on kernel method is proposed. First, the background of infrared image is predicted by KEWLS nonlinear regression algorithm, and then the target in residual image is detected by adaptive threshold. Nonlinear regression and infrared sequence image detection experiments show that the kernel method greatly improves the algorithm’s nonlinear function estimation and infrared background prediction ability.