基于局部多尺度低秩分解的红外小目标检测算法

针对红外低秩块模型计算复杂度大,容易误判等不足,提出了一种更加有效的红外小目标局部多尺度低秩分解检测算法。该算法首先利用非下采样金字塔变换对红外小目标图像做多尺度分解；接着,将分解出的高频子带进行融合,通过融合后的高频信息提取出目标感兴趣区域；最后,利用红外小目标背景的非局部自相关性质对感兴趣区域进行分块,并对各块进行重新排列构成一个新的矩阵；最后,对该矩阵做低秩分解,提取出红外小目标。实验结果表明,与其他低秩分解类方法相比,所提出算法速度更快,提取效果更好,是一种性能优越的方法。

Infrared small target detection algorithm based on local multi-scale low rank decomposition

Aiming at the disadvantages of large computational complexity and bad judgment of infrared low rank patch model,a more efficient multi-scale local low rank decomposition detection algorithm for infrared small targets is proposed.Firstly,the method uses the nonsubsampled pyramid transform to perform multi-scale decomposition on the infrared small target image.Secondly,the decomposed high-frequency sub-bands are fused,and the high-frequency information is extracted to extract the region of interest.Thirdly,the non-local autocorrelation property of the small target background is used to divide the region of interest and rearranges the blocks to form a new matrix.Finally,the matrix is decomposed into low rank to extract infrared small targets.Compared with other low rank decomposition methods,the proposed algorithm has better extraction performance and faster speed.The simulation results also prove its superiority and effectiveness.