联合行稀疏与l1/2-范数稀疏表示的红外小目标检测算法

Joint row and half-norm sparse representation algorithm for infrared small target detection

针对红外图像的小目标检测问题,提出了一种联合行稀疏与l1/2 -范数稀疏表示的小目标检测算法。首先采用滑动窗口对原始红外图像进行分块,构建分块图像矩阵;然后,根据稀疏与低秩表示理论建立行稀疏与l1/2 -范数稀疏表示的模型,并利用交替方向乘子算法求解得到稀疏矩阵和低秩矩阵;最后,通过图像重构,得到小目标的图像,并采用阈值分割的方法确定小目标的真实位置。实验结果表明,算法可以实现对不同背景红外图像中的小目标准确检测,与红外块图像建模算法相比,检测后的图像信杂比接近的情况下,检测速度提升了约1倍。

Aiming at the problem of small target detection in infrared images, a new algorithm for small target detection is proposed, which joints row sparse representation and 1 2 -norm sparse representation. First, the algorithm first uses sliding window to block the original infrared image and constructs the block image matrix. Then, according to the sparse and low rank representation theory, the model of row sparse and 1 2 - norm representation is established, and the sparse matrix and low-rank matrix are obtained by using the alternating direction method of multipliers. Finally, the target image is obtained by image reconstruction, and the real location of small target is determined by threshold segmentation method. Experimental results show that the algorithm can accurately detect small targets in infrared images with different backgrounds. Compared with the infrared patch-image model, the detection speed of the proposed algorithm is about twice as fast when the signal-to-clutter ratios of the detected images are close to each other.