【作者】 芦丽明；

【导师】 李言俊；

【作者基本信息】 西北工业大学 ， 导航、制导与控制， 2002， 博士

【摘要】 随着现代战争的发展，对精确制导信息处理技术的要求越来越高，ATR(自动目标识别)技术已成为精确制导技术发展的方向，而ATR技术的发展则是起源于人类对生物视觉的模仿。基于这一点，本文在对复眼的研究及讨论的基础上，提出了一些自动目标识别的技术。复眼作为一个现实中的生物的视觉系统，它首先具有一般生物眼睛的视觉系统功能，在此之外，还拥有其独特的特点。本文就是从这两点展开讨论研究。首先，把复眼作为一般视觉系统看待，从图像的预处理、图像的分割、特征提取以及目标的分类与识别，这一系列的一般视觉处理过程来考虑。然后，根据对其复眼特有的特点的理解，提出了一种仿复眼的多模复合制导技术。本文所作的工作主要在以下几个方面： 1、红外图像的预处理：首先介绍了传统的图像预处理方法，然后重点介绍和总结了前人关于侧抑制机理在红外图像预处理中的应用。其中包括侧抑制理论的数学模型、稳定性判据、时域特征、频域特征、侧抑制网络突出图像边框、增强反差的功能以及在图像分割中的应用。 2、红外图像的分割：首先系统地介绍了图像分割的模型和分类，然后力求从图像分割的自适应和鲁棒的角度，仔细讨论了最佳熵和模糊熵的图像分割技术，引入了遗传算法的寻优规则。最后又提出一种避免求阈值的多目标的图像分割算法，该算法简单可靠，具有一定的自适应性和鲁棒性。 3、红外图像的特征提取：先介绍了传统的特征提取的方法，然后提出了一种多通道特征提取的方法，包括了基于小波的频率多通道和基于灰度的多通道的特征提取方法。基于小波的频率多通道是从小波多分辨分解的角度，提取各频率通道的灰度特征形成特征向量，该方法可以很好的区分目标，但不具有旋转不变性。因此随后提出了基于灰度多通道的特征提取方法，该方法是把目标灰度分成几个区段，然后在各灰度区间分别提取目标的灰度特征，形成特征向量。该方法克服了频率多通道特征提取的缺点具有旋转不变性。 4、红外目标的分类与识别：由于目标特征提取中，受到各种因素的影响，提取的特征不可避免的带有噪声，即是具有一定的不确定性。本论文则主要研究了如何根据具有不确定性的目标特征向量来识别目标。首先研究了D-S证据推理理论，和D-S证据理论中的合成规则，Yager合成规则的缺点。然后提中文摘要出了一种基于多特征的目标识别方法，其中包括初始概率赋值的确定方法和一种新的合成规则。利用该合成规则，可以避免D一S合成规则和Yager合成规则不能很好解决的证据冲突时的合成。 5、仿复眼多模制导:首先在基于对复眼的结构和功能研究理解的从础「，提出了一种对视觉系统的观点:不需要两只复眼，单只复眼就可以得到日标的三维信息。在此理解的基础上，提出了一种多模导引的框架:两个点源探测头和一个成像探测头。一个点源头模拟了复眼中的一个单眼，成像头模拟了整只复眼对环境和目标的成像，利用两个点源头的信息可以得到日标的三维空间方位信息。在最后，研究了这种多模框架的多传感器对目标的检测融合问题。更多还原

【Abstract】 With the development of war, the requirement on information processing techniques for precision guidance is becoming higher and higher. ATR has become very important, this technique imitate human visual system. So this paper does certain researches on ommateum and some ATR techniques are proposed in this paper. As a real visual system existing in nature, ommateum not only have the same function as other visual systems, but also have its special feature. So research in this paper on ommateum contain two points, first researching ommateum like other normal visual system, considering contents including preprocessing of IR images, target image segmentation, feature extraction and target classification; second researching ommateum’s special feature. Based on researching on ommateum"s special feature, a multi-mode guidance method is proposed in this paper. The main achievements of this paper are as follows:1 , The preprocessing of IR image: first introducing traditional methods on preprocessing of IR image, second emphasizing on introducing and summarizing lateral inhibition theory and application that predecessors had researched. This contains lateral inhibition theory’s mathematics model, stability criterion, time domain and frequency domain feature, the function of enhancing the image’s contrast and the object’s frame and other functions applying in image segmentation.2 , Segmentation of the IR image: First systemic introducing image segmentation’s model and classification, second emphasizing on researching image segmentation using best entropy or fuzzy entropy based on genetic algorithm . last a method for multi-object image segmentation avoiding searching for threshold is proposed, that is simple, stability and has certain self-adaptive and robust.3, Feature extraction: first introducing traditional feature extraction methods, then multi-channels feature extraction methods is proposed. One method is multichannels based on. frequency using wavelet theory, another is multi-channels based on gray. The first method is based multi-scale decompression using wavelet theory, it extract gray feature from each channel to form feature vector, it can distinguishobjects, but this method not have invariant property when object rotate. So the second method is proposed, it has not the same shortcoming, the method is based gray multi-channel. In this method, gray is partitioned into gray regions and feature is extracted in each regions and features form vector.4, IR object’s classification and recognition: because of various factors in the process of feature extraction, the feature extracted is polluted certainly by noise, that the feature has certain uncertainty. This paper emphasizes on researching how to classify object based on feature vector having uncertainty. First researching on D-S evidence theory, .Combination Rules of Evidence Theory, Yager combination rules and this theory’s shortcoming. Then a new method for object classification based on multi-feature is proposed, in this method basic probability assignment and combination rule is resolved. When the evidences have conflicts, D-S combination and Yager combination can’t resolve this thing, the new method can resolve this conflict question.5 , Multi-mode guidance: Firstly, a idea about visual system is proposed based on researching on ommateum’s structure and function. The idea is that: one ommateum can get object 3-D information in the space, not using two ommateum. Based on this point, a multi-mode guidance method is proposed, this method contains two point-detectors and one image detector. One point-detector imitates facet in ommateum, image detector imitates one whole ommateum, and it’s visual function. Object’s 3-D information is got using two point-detectors. In the last, the object detection question about multi-sensor’s data fusion is researched.更多还原