空间太阳望远镜稳像系统中图像相关器的研究

【作者】 李长松；

【导师】 金声震； 姜爱民；

【作者基本信息】 中国科学院研究生院（国家天文台） ， 天文技术与方法， 2008， 博士

【摘要】 空间太阳望远镜是我国自主研发的第一颗天文卫星。主要用于对太阳的全日制、多波段以及高分辨率观察。将有助于太阳小尺度磁场和活动区磁场精细结构的研究,并有望取得突破性进展,因此曾被国外媒体称为“雄心勃勃的大计划”。空间太阳望远镜的1m主光学系统具有0.1″的瑞利衍射极限分辨率。望远镜本身的卫星平台仅能提供±6″的指向精度和3″/s的姿态稳定度,由于卫星本身的震动以及太阳自转会造成积分时间内的成像模糊。在必需增加积分时间以提高CCD图像信噪比的情况下,需要利用相关跟踪器改正,实现对成像的运动补偿。本文对空间太阳望远镜相关跟踪器的目标图像特征进行了分析。考虑到太阳米粒组织的对比度比较低、形状不规则等特点,沿用了国内外一贯采用的相关算法实现图像偏移检测。相关算法在整像元偏移检测上具有很好的效果,辅助亚像元拟合可以得到更高的检测精度。本文通过对大量米粒图像进行仿真分析发现,引入修正参数后的曲面拟合算法可以进一步提高亚像元偏移检测精度。其中修正参数是利用统计学方法对其分布进行分析后,根据数学期望的估计值得到的。对于整个相关跟踪系统来说,相关计算十分费时。过去由于器件水平有限,太阳望远镜的相关跟踪系统难以实现快速计算,造成了系统具有明显时滞特征。而纯时滞系统的闭环特性与时延有反比关系,因此寻求快速相关实现方法成为类似系统所需解决的关键之一。快速相关器的设计可以利用简化相关算法实现,主要是减少乘法的次数,例如以减法相关算法取代乘法相关算法。但是,减法相关算法容易产生零漂。本文在总结前人的经验基础上,根据当今集成电路发展的水平,采用了高集成度的FPGA实现相关运算。利用并行算法将相关运算进行分解,并与FPGA的结构特点相结合,设计出并行相关计算单元。使之具有模块化、数控分流等特点。充分利用了硬件资源,达到缩短计算时间的目的。利用FPGA实现相关运算的数据结构为浮点格式。考虑到资源有限,在保证运算过程不发生溢出的情况下,采用18 bits的浮点运算。通过实际测试,相关矩阵的计算结果与双精度浮点的相比,相对误差在2‰～6‰之间。利用理论推导、仿真波形和示波器测试都可以验证,整个相关计算时间约为240us。满足空间太阳望远镜的需求。综上所述,拟合修正参数的引入改善了亚像元偏移检测精度,为高精度亚像元偏移检测提供了算法保证;利用并行算法和FPGA结构特点相结合的方法,极大缩短了相关运算的计算时间。实现了计算时间和计算精度指标的优化。本文的相关跟踪计算单元不但满足了相关跟踪器的指标需求,而且为其他类似实时高精度数据处理系统提供设计参考。更多还原

【Abstract】 The Space Solar Telescope (SST), proposed by Chinese scientists independently, is described as "the ambitious big plan" by the overseas media. Its main objective is to study the sun using five payloads, which will be helpful in researching the small-scale solar magnetic field, and hope to achieve a breakthrough.On the one hand, we need to increase the integration time to improve the signal to noise ratio (SNR) of the CCD image. On the other hand, the satellite platform of SST can only provide±6" pointing accuracy and 3"/s attitude stability. So the Main Optical Telescope (MOT) can not realize its 0.1" resolution without help. Under this circumstance, it is necessary to use a correlation tracker (CT) to stabilze the image.In this paper, considering the low contrast and irregularity of solar granulation, we use the correlation algorithm to detect the image movement, which is widely adopted by both foreign and domestic scientists. The correlation algorithm can be calculated by three steps. Firstly, the integral offset between the reference image and the live image can be found by getting the coordinates of the peak of correlation matrix. Secondly, sub-pixel offset can be determined by fitting the 5 by 5 matrix around the peak of correlation matrix with a parabolic surface. Finally, the total offset can be calculated by adding the integral offset to sub-pixel offset. After a large number of simulation experiments, we find that adding two parameters, obtained by statistical method, at the final step can reduce the error in image offset detection.It is a pure time delay system characteristic that its bandwidth is inverse proportion to the delay time. So, to realize the correlation algorithm as far as possible is one of the correlation tracker’s key tasks. By full utilizing the characteristics of FPGA, we realize the design of correlation tracker’s calculation unit and short the calculation time in great degree by modular design and pipeline organization.The 18-bit floating-point data structure can not only satisfy the restriction of XCV800 FPGA resource, but also ensure no overflow occurs during the caluculation. Through many tests, the relative error of correlation matrix between the results of our design and MATLAB program is better than 6‰.From theoretical analysis, simulated waveform, and oscilloscope’s measurement, we can draw the conclusion that the correlation calculation time of our design is about 240us, which is great shorter than other designs.Above all, we improve the accuracy of image offset detection by using the modified fitting algorithm, and reduce the correlation calculation time by combining the parallel algorithm and the characteristics of FPGA. The solution we provided can not only meet the needs of correlation tracker, but also provide a reference to other similar applications.更多还原