利用不变矩去除恒星的空间小目标检测方法

Space small targets detection using moment invariants removing stars

相邻帧配准差分去除恒星方法,在处理16位灰度图像时,由于高亮度恒星的插值误差大于目标灰度与背景之差,导致差分不能有效去除恒星干扰。针对该问题,引入模式识别中的不变矩概念,提出基于不变矩的空间小目标检测方法。该方法从恒星之间的相对位置入手,避开通过灰度差分消减恒星的过程。利用相邻帧中恒星不变矩相等的特性,寻找三颗恒星作为参考恒星组;向参考恒星组逐一加入未知星点,通过判断增加星点后不变矩是否仍相等来提取恒星;利用目标运动的连续性和一致性从剩余星点中提取目标。通过对仿真恒星点配准差分说明问题的由来,利用实摄序列图像对配准差分法和不变矩方法对比实验,结果表明,配准差分法无法去除的恒星可以通过不变矩方法去除,该方法能够从16位灰度图像中有效检测出小目标。

Registration and difference can＇t exclude stars when processing the 16-bit gray images because interpolation error of high light star is larger than the difference of target and background.Focus on this problem, a detection method of space small targets is presented based on moment invariant which is a concept of pattern recognition.This method starts from relative position of stars and avoids from removing stars by gray difference.It finds three stars as reference star group used the character that star moment invariants are equal in neighbor frames.The unknown points are joined to reference star group one by one and stars are extracted by judging whether the moment invariants of group are still equal after inserting new point.Targets are extracted from residual points using the continuity and consistency of targets movement.The problem is illustrated by registation and difference of emulation stars.The contrast experiments between registation and difference method and moment invariants method used real sequence images are processed.Results indicate that stars can be wiped off by this method which can＇t be removed by registation and difference and the method can detect small targets from 16-bit gray images efficiently.