基于随机化张量算法的红外弱小目标检测  

Infrared small target detection based on randomized tensor algorithm

作　　者：蹇渊 黄自力[1] 王询 JIAN Yuan;HUANG Zili;WANG Xun(Southwest Institute of Technical Physics,Chengdu 610041,China)

机构地区：[1]西南技术物理研究所,成都610041

出　　处：《激光技术》2024年第1期127-134,共8页Laser Technology

摘　　要：为了降低基于张量低秩稀疏分解的红外弱小目标检测算法的计算复杂度,提升红外弱小目标的检测性能,将图像时空张量与随机化算法进行结合,提出了一种基于随机化张量算法的红外弱小目标检测算法。首先将红外图像序列构造成时空张量作为张量优化模型的输入,然后使用随机化张量算法求解张量优化模型,最后将计算得到的稀疏张量还原为图像,获得目标图像。结果表明,相比于传统基于低秩稀疏分解的算法,所提出的算法不仅计算速度快,而且具有较好的弱小目标检测性能。该研究为提升基于张量低秩稀疏分解的红外弱小目标检测算法的运算速度提供了参考。In order to reduce the computational complexity of the infrared small targets detection algorithm based on tensor low-rank sparse decomposition and improve the detection performance of infrared dim targets,an infrared small target detection algorithm based on the randomized tensor algorithm was proposed.The algorithm combines the spatial-temporal tensor of the image with the randomized algorithm.Firstly,the infrared image sequence was constructed into spatial-temporal tensors as the input of the tensor optimization model,and then the randomized tensor algorithm was applied to solve the tensor optimization problem.Finally,the target image was obtained by restoring the calculated sparse tensor to the image.The results demonstrate that compared with the traditional algorithm based on low-rank sparse decomposition,the proposed algorithm is faster and also has good detection performance.This study provides a reference for the algorithm acceleration of infrared small target detection based on tensor low-rank sparse decomposition.

关 键 词：图像处理 低秩稀疏 红外弱小目标检测 随机化张量算法 时空张量 

分 类 号：TN911.73[电子电信—通信与信息系统]