【作者】 李淼；

【导师】 高慧斌；

【作者基本信息】 中国科学院研究生院（长春光学精密机械与物理研究所） ， 机械电子工程， 2012， 博士

【摘要】 光电经纬仪是用于光电跟踪测量的重要设备，跟踪性能是评价光电经纬仪的主要指标之一。理论上，光电经纬仪的跟踪性能指标是以等效正弦为基础而提出的，而目前室内却难以产生纯的等效正弦信号。因此，实际室内检测中，通常是采用光学动态靶标模拟的运动目标进行检测。然而，理论分析发现，光学动态靶标模拟的运动目标投影到光电经纬仪的方位和俯仰两个方向上的运动规律都不是纯正弦运动，而是包含大量高次谐波的正弦波的合成。特别是在靶标架设的高度比较高，或者靶标运动周期较短（T <2π）时，这种畸变就更加明显，偏离正弦的程度就越严重。而且由于正割补偿效应的存在，这种偏离还表现出方位比俯仰更明显的特点。因此，导致利用光学动态靶标检测时，输入条件与指标要求不完全对应，对光电经纬仪跟踪性能的评价也不够客观准确。本文针对如何避免动态靶标高次谐波影响，更加客观评价经纬仪跟踪性能开展了以下研究工作：1．分析了用动态靶标检测评价经纬仪跟踪性能的局限性，对光电经纬仪运动的动力学特性，结构动态特性和非线性因素进行了分析，说明了理论建模难以准确描述光电经纬仪全部动态特性的原因。2．提出了一种实际数据与计算机仿真模型结合的经纬仪跟踪性能的评价方法，通过充分利用电视系统对光学动态靶标的跟踪信息，建立跟踪伺服系统误差模型的方法，采用与技术指标要求相对应的等效正弦信号，实现了输入信号与指标要求信号一致，从而使对光电经纬仪的评价更为合理。3．通过神经网络（RBF，GRNN）方法和机器学习方法（SVM，ELM）建立了经纬仪的跟踪误差模型。根据光电经纬仪跟踪靶标得到的实际数据进行网络的学习和训练，在一定程度上克服了经纬仪内部的非线性因素和难建模动态，保证了所建模型能够在最大程度上接近实际系统。对不同建模方法进行对比，选定了适合的模型，并进行了模型验证，给出了GRNN和ELM更适合用于经纬仪建模的结论。4．将根据技术要求指标计算得到的计算机模拟的等效正弦信号，输入到所建的网络模型中，得到在等效正弦输入下的跟踪误差，有效地分离了动态靶标高次谐波影响，更加客观地评价了经纬仪的跟踪性能。5.采用2组仿真数据和建模效果较好的两种方法（GRNN，ELM）验证了本文提出的方法的有效性。采用2组可见电视，1组红外电视跟踪的真实数据，重新实验，进一步验证了本文提出的方法。真实数据的实验结果表明，本文提出的方法是可行而有效的。更多还原

【Abstract】 The opto-electrical theodolite is an important device for opto-electrical tracking measurement,and the tracking performance is one of the most important evaluation indexes. In theoretical, thetracking performance of the photoelectric theodolite based on equivalent sine theory, but this kindof signal is hard to emit indoor. So in practice, we usually use opto-dynamic target to simulate theobject to complete the measurement. According to the theory analysis, we known that themovement that opto-dynamic tartget simulated projected to photoelectric theodolite’s azimuthdirection and elevation direction are not pure equivalent sine, and it synthesized many high orderharmonic sine wave. If we use dynamic target to measure tracking performance of thephotoelectric theodolite, the input signal is not completly meeting requirement index. So it isunreasonable to use the opto-dynamic target to measure and evaluate and tracking performance ofthe photoelectric theodolite.In this paper, we focusd on this problem and did some researches on how to avoid theinfluence from opto-dynamic target high order harmonic and how to evaluate the photoelectrictheodolite tracking performance more reasonable and objective. The paper involved the followingpart:1. The paper analyzed the reason why it is unreasonable to use the opto-dynamic target tomeasure and evaluate and tracking performance of the photoelectric theodolite. It is also analysedthe dynamic characteristic, structure dynamic as well as nonlinear factors in the photoelectric theodolite. All of these revealed that it is hard to build the photoelectric theodolite model viatheoretical model method.2. The paper put forward a new kind of method that based on the real data as well as appliedthe computer simulation model to evaluate the theodolite’s tracking performance. This methodmade full use of the tracking information from the TV system, and applied the proper equivalentsine signal which computed from the index requirement, which realized the input signal meet therequirement index and avoided the high harmonic from opto-dynamic target and reduce itsinfluence on photoelectric theodolite tracking performance evaluation.3. Built the tracking error model of the photoelectric theodolite by neural network method(RBF, GRNN) and machine learning method (SVM, ELM).Applied the real data from photoelectric theodolite tracking the dynamic target to train thenet, which can partly get over the nonlinear factor as well as hard build dynamic in the theodolite,and make sure that the model can fully near the actual system.Compared and verified different model method, and arrived conclusion that GRNN and ElMare more suitable for photoelectric theodolite model.4. Put the equivalent sine computed from index requirement to the built net model, and getthe tracking error that filter the high harmonic influence from opto-dynamic target and evaluatethe tracking performance of the theodolite more reasonable.5. Use two group of simulation data and two effective model method (GRNN, ELM) testifiedthe evaluation method putting forward from this paper.Use two group of visiable TV and one group of infrared TV real tracking data to do theexperiment and the result revealed that the method that put forward in this paper is useful andpractical.更多还原