复杂背景下的空间目标自动识别技术

发布时间：2018-01-01 11:09

本文关键词：复杂背景下的空间目标自动识别技术　出处：《中国科学院长春光学精密机械与物理研究所》2017年博士论文　论文类型：学位论文

【摘要】：随着人类对空间技术开发利用规模的不断扩大,空间已经成为获取和传递信息的重要平台,因此,对空间的控制尤为重要。空间控制的前提是对空间目标的监视,包括空间目标探测识别和跟踪。空间目标主要指空间碎片和人造卫星,及进入地球外太空的各种飞行物。空间目标的精确识别与定位是空间目标监视的主要目的,是空间态势感知的重要组成部分,也是确保载人航天安全和空间预警的重要技术保障。所以对空间目标探测技术的研究非常具有现实意义和应用价值。本论文采用的是地基光电探测系统对空间目标进行探测和跟踪。与雷达和天基平台相比,这种探测方式的测量精度高,直观性强,成本低,不受地面杂波干扰的影响,但容易受到天气变化的影响。为兼顾目标搜索效率和探测能力,一般采用大视场光学望远镜对远距离空间目标进行搜索测量,进而达到空间目标快速发现和识别的目的。地基探测存在几个难点:一是大视场光学系统本身存在光学畸变,背景不均匀,同视场恒星过多,图像处理实时性要求高;二是空间目标尺寸小、距离远,亮度受太阳照射方向影响变化大,信噪比低;三是空间目标受轨道高度差异影响、目标速度和运动差异较大。这些问题,给自动识别和处理带来很多挑战,因此如何在复杂背景条件下自动识别出更多,更暗,更弱目标是本论文的主要目标。本课题主要从空间目标测量光电望远镜系统、星图图像处理、空间目标检测识别和空间目标定位几个方面入手,研究提升空间目标识别与跟踪能力,提高空间目标定位精度的方法。具体完成研究工作如下:(1)论文首先详细介绍了对空间目标进行搜索测量的光电望远镜系统的基础知识,包括望远镜的光学系统、机械结构以及CCD探测器等。(2)详细介绍了星图图像的处理技术,包括星图的去噪和阈值分割技术,以提高星图信噪比。在分析了星图成像和椒盐噪声特性的基础上,提出了“基于能量函数的极值中值滤波星图去噪算法”,该算法针对疑似噪声点采用二次检测的方式,并且结合改进的自适应中值滤波和能量函数模型进行灰度值恢复。客观评价中,图像信噪比PSNR最高可提高3倍多,均方误差MSE减小为含噪图像的3.16×10-5。实验结果表明,该方法可有效地降低传统方法的噪声误检问题,同时提高噪声图像的恢复精度,很适合去除星图的椒盐噪声。(3)论文在分析空间目标运动特性的基础上,针对空间目标和恒星背景之间存在的运动特性差异,充分利用空间目标在时域上的分布信息,同时结合其空域相关性,提出了“基于运动信息的星图暗弱空间运动点目标检测算法”,引入相关系数矩阵,成功提取目标运动轨迹,最后给出运动速度估计模型。试验结果表明,所提方法能够在保持较低的虚警概率下获得较高的检测概率,优于其它比较方法,与单纯扩大望远镜口径相比,该方法为提高空间暗弱目标识别能力提供了具有更高性价比的有效途径。针对传统星图空间运动目标检测方法对星图帧间图像亮度,帧间配准以及成像模式等具有较高的敏感性的缺点,根据星点间拓扑结构的稳定性,提出了“基于距离矩阵的星图运动目标检测方法”,该方法利用稳定的星点几何结构信息完成运动目标的检测,对星图帧间图像亮度,观测平台抖动,帧间失配以及成像模式等具有高鲁棒性。仿真试验和真实数据试验表明,该方法在帧间失配等情况下,能够从背景恒星中有效地识别空间运动目标,并保持较低的虚警率。通过构建星图时谱图,在此基础上,分析星图中主要对象(目标、恒星和背景)的时域特征,并利用目标的时域信号的单脉冲特征实现运动目标的检测,提出了“一种基于时域特征的空间运动目标检测方法”。真实星图目标检测试验表明,此方法能够有效检测星图序列中运动目标轨迹,在虚警率为8×10-5时,检测概率可达99%,优于传统空间目标检测算法。(4)通过分析传统质心提取算法的不足,论文提出了“基于各向异性的高斯曲面拟合的星点质心提取算法”,提出了各向异性的高斯曲面拟合模型,该模型通过使用两个不同的高斯模糊参数和旋转因子,可以捕捉空间目标不同方向的各向异性特征,适合卫星由于运动造成的随机方向模糊。仿真实验和真实数据实验表明,该方法的总体定位精度可分别达到0.008和0.04,能够准确提取星图目标的质心,较传统方法有较大提高。
[Abstract]:As the development and utilization of space technology continues to expand the scale, space has become an important platform for acquiring and transmitting information, therefore, control of space is particularly important. The premise of space control is the space surveillance, including space target recognition and tracking. Space target mainly refers to the space debris and artificial satellites, and into the earth outer space flying objects. Accurate positioning and recognition of space targets is the main purpose of space surveillance, is an important part of space situational awareness, but also to ensure the important technical support of manned spaceflight and space warning. So research on space target detection technology has very realistic significance and application value. This paper uses is the foundation of the photoelectric detection system for space target detection and tracking. Compared with radar and space-based platforms, the accuracy of detection is high Intuitive, strong, low cost, not affected by ground clutter, but is easily affected by the change of the weather. In order to balance the target search efficiency and detection ability, generally adopts a large field optical telescope for space target search distance measurement, so as to achieve rapid discovery of space targets and identification purposes. There are several difficult foundation a: there is optical distortion of wide-angle optical system itself, the background is not uniform, the same star too much, real-time image processing; two is the space target size, distance, brightness by solar irradiation direction changes, the signal-to-noise ratio is low; the three is affected by the space target orbit height difference. The target speed and movement are different. These problems bring many challenges to the automatic recognition and processing, so how in the condition of complex background automatically identify more and more dark, more weak target is the main To the target. This topic mainly from the space target measurement of photoelectric telescope system, star image processing, several aspects of space target detection and target location of research to improve target recognition and tracking method to improve the ability of space, space target positioning accuracy. The finished research works are as follows: (1) this paper first introduced the basic knowledge search system measurement optoelectronic telescope for space target, including the telescope optical system, mechanical structure and CCD detector. (2) introduces the processing technology of star image, denoising and threshold segmentation technology including chart, chart in order to improve the signal-to-noise ratio. Based on the analysis of the star image and noise the characteristics of the proposed denoising algorithm, "star extremum median filtering based on the energy function, the algorithm uses the two detection methods for suspected noise, and Combining the adaptive median filter and energy function model of improved gray value recovery. The objective evaluation of image SNR PSNR maximum increase more than 3 times, the mean square error decreases to 3.16 MSE * 10-5. experimental results of noisy images show that the method can effectively reduce the noise of the traditional method of error detection, at the same time to improve the noise image restoration precision, is suitable for the removal of salt and pepper noise map. (3) based on the analysis of the target motion characteristics of space, according to the difference in motion between space target and the background of the stars, make full use of the space distribution information of target in the time domain, combined with the spatial correlation, put forward the "space motion the faint star point motion information of the target detection algorithm based on the correlation coefficient matrix, the successful extraction of target trajectory, velocity estimation model is given. The experimental results show that the proposed party Method to obtain high detection probability while maintaining low false alarm probability, is better than other methods, compared with the simple expansion of the telescope, the method provides an effective way to improve the cost-effective space faint target recognition abilities. Detection method for traditional star space moving target in terms of image brightness, the sensitivity of the interframe registration and imaging mode has higher shortcomings, according to the star topology structure stability, proposed the "star moving target detection method based on distance matrix, the detection method using star geometry information of steady motion target, in terms of image brightness, observation platform frame jitter. Mismatch between imaging mode and high robustness. Simulation results show that the test and real data, the method in the inter frame mismatch case, from the background star In the effective identification of space object, and maintain a low false alarm rate. By constructing the star spectrum diagram, on this basis, the main object of analysis chart (target, star and background) time domain characteristics, detect and use single pulse time domain signal characteristics of the target moving target, put forward "a moving target detection method based on time domain feature. Show the true star target detection test, this method can effectively detect the moving target trajectory map sequence, the false alarm rate is 8 * 10-5, the detection probability reaches 99%, is superior to the traditional space target detection algorithm. (4) through the analysis of the traditional centroid extraction algorithm insufficient, the paper proposes a" star centroid extraction algorithm of anisotropic Gauss surface fitting based on the proposed Gauss surface fitting model of anisotropy, the fuzzy parameters of the model by using two different Gauss And the rotation factor and anisotropy can capture the spatial objects in different directions, suitable for satellite due to random motion blurred direction. Simulation results and real data, the overall positioning precision of this method can reach 0.008 and 0.04, can accurately extract the target centroid chart, is greatly improved compared with traditional methods.

【学位授予单位】：中国科学院长春光学精密机械与物理研究所
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】