摘要
为了从序列图像中检测出空间小目标,提出了一种改进的动态规划算法。介绍了动态规划小目标检测算法的发展过程。针对工程应用中为了获取亚像素质心而对图像散焦处理的情况,对原算法的评价函数递归方程进行改进,提出用方向加权的多点能量累积代替原算法的单点累积,并将多速度平面分别计算的方式简化为速度初始化与历史速度修正方式。最后,针对算法实现过程中遇到的初始状态计算、恒虚警阈值以及轨迹数据结构等关键问题进行了说明。在实验参数条件下,计算量比原算法减少了约50%。实摄图实验结果表明,方向加权的多点累积算法5帧即可检测出目标,而原算法在第10帧时仍有大量虚假轨迹;在第5帧的20条最大评价值轨迹中,多点法的评价值信噪比比原算法提高42%。方向加权的多点累积算法可以有效抑制孤立噪声点产生的虚警,提高算法的检测能力。
An improved dynamic programming algorithm was presented to detect space small targets from sequence images. Firstly, the development and current situation of the dynamic programming algorithm used in small target detections were introduced. As the defocus was used to process commonly a project to obtain the sub-pixel center of a target,the recursive equation of score function was improved. A multi-point accumulation algorithm with direction weight was presented to replace the single-point accumulation in an original algorithm, and the original multispeed plane calculation was also simplified. Finally, the key processes including state initialization, constant false alarm rate threshold and track data structure were discussed. The analysis indicates that the computational complexity has reduced about 50% as compared with that of original algorithm. Experimental results also indicate that targets can be detected in 5th frame with the multi-point algorithm, while a number of fake tracks are found until the 10th frame with the original algorithm. Furthermore,the score SNR of 20 tracks with maximum score in the 5th frame of multi-point algorithm is 0.87 higher than that of original algorithm. The multi-point accumulation algorithm with direction weight is able to reduce the false alarm caused by a speckle noise efficiently and to raise the detection ability of dynamic programming.
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2010年第2期477-484,共8页
Optics and Precision Engineering
基金
国家863高技术研究发展计划资助项目(No.2006AA703213D)
关键词
动态规划
小目标检测
多点累积
空间图像
dynamic programming
small target detection
multi-point accumulation
space image
