【作者】 刘立峰；

【导师】 陆明泉； 马惠敏；

【作者基本信息】 清华大学 ， 信息与通信工程， 2010， 硕士

【摘要】 对于寻的飞行器来说,空中目标抗干扰识别和跟踪一直是一个非常重要的问题。飞行器从发射到捕获目标,光学远场探测器和近场探测器先后工作,两个阶段所面临的干扰是不一样的。本文主要针对红外远场探测器和激光成像近场探测器的抗干扰识别问题展开研究。对于红外远场探测器来讲,其所面临的主要干扰是目标向不同方向同时抛出的多个燃烧干扰源。本文首先实现了一个远场探测器电信号模拟器,可以利用计算机仿真的数据,模拟生成目标和干扰源的红外探测信号,从而为远场探测器抗干扰识别跟踪算法提供了数据。本文在所设计的抗干扰硬件平台的基础上,根据远场探测器发挥作用的全过程和机器学习得到的结果,通过对各个状态的控制,复合使用波门技术、航迹记忆和分析、脉冲分析技术实现了对正确目标的识别。实测结果显示,本文算法具有良好的抗干扰性能。对于激光成像近场探测器,其所面临的主要干扰是云雾。空间获得云雾图像代价高昂,因此,通过计算机仿真获得云雾图像就显得有必要,本文在所建立的云雾模型基础上,尝试用蒙特卡罗仿真方法获得了云雾回波图像,并通过文献调研和实验对其进行了验模。针对激光成像近场探测器获得图像高噪声、多姿态的特点,本文提出了一种基于Zernike矩的目标识别算法,相较于点和线特征,Zernike矩特征是一种更为稳定的特征,本文给出了其硬件实现的具体过程,同时给出了图像放缩、简单分割以及多尺度目标识别的硬件实现过程。对算法进行测试显示,本文算法不仅可以获得一个更高的识别率,同时也可以获得一个更低的误识率。本文通过对红外远场探测器抗干扰算法和激光成像近场探测器的综合研究,极大地提高了寻的飞行器命中目标的概率,有很重要的理论和实用价值。更多还原

【Abstract】 For aerial vehicle, the aerial target identification has been a very important issue. The optical far-field detector and near-field detector have been working from the vehicle launching to capture the target. The interferences of the two stages are not the same. This issue is written mainly for the target identification issues from interferences of the third-generation infrared far-field detector and laser near-field imaging detector.For the third-generation infrared far-field detector, the interferences they face are the interferences thrown by the target in different directions. First, this issue implements a far-field detector signal simulator, using the data generated by computer simulation to simulate the detected infrared signals of targets and interferences and to provide the data for the recognition algorithm of the far-field detector. And then, this issue achieves the correct target identification on the hardware platform designed based on the whole process of the far-field detector working and the results of machine learning, by controlling of each state, combined with wave-gate technology, track memory and analysis, pulse analysis techniques. Experimental results show that the algorithm has good robustness.For the laser near-field imaging detector, the main interference it faces is cloud. Because acquiring images of cloud is so expensive, using the computer simulation to obtain the cloud images becomes necessary. This issue tries using the Monte Carlo simulation method to obtain the cloud echo images based on the cloud modeled, and tests the model through literature research and experiments. For the high noise and multi-profile images laser near-field imaging detector obtain, this issue presents a target recognition algorithm based on Zernike moments which are more stable characteristics comparing with point and line features. A hardware implementation of Zernike moments calculation is given, and also the process of the image scaling, an simple image segmentation and a multi-scale object recognition. The algorithm tests show that method in this issue can obtain not only a higher recognition rate, but also a lower error rate.Based on the comprehensive study of far-field detector and laser near-field imaging detector anti-interference target identification, which greatly improves the probability of the vehicle hitting the targets, this work has an important theoretical and practical value.更多还原