【作者】 王国锋；

【导师】 李言俊；

【作者基本信息】 西北工业大学 ， 导航、制导与控制， 2003， 博士

【摘要】 红外成像制导已成为当今精确制导武器的重要技术手段。但目前的目标图像探测技术，无论是光机扫描成像还是凝视焦平面成像，探测的角度都有限，难以用一个探测器实现360°范围内的探测，因此难以实现导弹全方位发射和攻击。在此背景下，蝇的复眼给了我们一个极好的启迪。本文针对蝇视觉在成像制导应用中需要解决的问题，如蝇复眼运动检测器阵列的算法及实现、图像的仿复眼镶嵌整合、蝇复眼目标检测机理、蝇仿生复眼的结构及在导弹上的配置等，集中研究了以下几方面内容： 1．蝇复眼运动检测器及图像运动速度估计 研究了初级运动检测器和运动检测器阵列的算法、特征及空间整合。推导出基于图像频谱密度的响应公式。研究表明：运动检测器可以精确检测多频谱自然图像速度，在运动检测器前增加生理学预滤波环节可以改善图像运动速度的估计。此结论极大地丰富了Reichardt初级运动检测器理论。探讨了运动检测器的工程实现电路，其实验结果与理论、蝇视觉实验结果非常吻合。 2．图像的仿复眼镶嵌整合技术 推导了基于空间坐标投影的图像配准算法，针对多幅静态图像提出了仿复眼的基于小波多分辨分析的图像镶嵌整合算法，建立了一种新的图像整合规则；针对具有运动目标的序列图像，提出了基于特征的仿复眼图像镶嵌整合算法。两种方法克服了现存的图像拼合算法只能得到局部最优解的缺点，对图像样本的要求大大降低，而且可以得到全优解。 3．仿蝇视觉的红外图像处理方法 研究了蝇视觉图形—背景分辨系统及其相应的算法，探讨了基于复眼侧抑制网络的图像预处理方法。提出一种仿蝇视觉目标检测的工程实现方法—基于模糊熵的图像分割方法，它分割速度快，简化了运算量，提高了运算速度。 4．仿蝇复眼的红外图像目标识别 提出一种基于红外目标特性和边缘不变矩的特征向量提取方法，提出了仿蝇视觉基于支持向量机的目标识别方法，该方法识别率比传统K近邻法高，且识别速度快。 5，蝇视觉系统在成像制导中的综合应用 提出了一种仿复眼，具有全方位视场的红外成像导引头物理模型，该导引头中文摘要模拟了整只复眼对环境和目标的成像，可以实现36了搜索视场。提出了相应图像简化配准算法，给出了蝇视觉系统在成像制导上的综合应用流程，为后续蝇复眼在成像制导上的具体应用打开了思路。更多还原

【Abstract】 Today, infrared imaging guidance has been one of the most precise guidance techniques. But today’s target imaging detecting techniques, including scanning imaging and gazing-array imaging, is difficult to detect 360?range using one detector, and it is difficult to launch and attack with full of orientation. Under this background, fly’s compound eyes give us a good edification. Aiming at some problems about fly vision applying to imaging guidance, such as, algorithm and realization of motion-detectors array, image mosaic technique of imitating compound eyes, target detection of compound eyes, realization of compound eyes in missile, and so on, the main contents of this paper are as follows:1 . Motion-detector of fly’s compound eyes and determination of velocity of moving image.This paper describes the algorithm of elementary motion-detector and motion-detector arrays, including the characteristic and space conformity. The formula based on spectrum density is given. It discusses motion-detector response to narrow-band image motion and to broadband images. Studies show that the motion-detector can provide velocity estimates in organism’s natural visual environment, and physiological prefilters can improve determination of velocity of moving image. This conclusion is affluent in the theory of Reichardt EMD, and it is the complementarity of HMD theory. This paper also discusses the realization of motion-detector, the circuit is given, and experiment result accords with theory and fly vision.2. Image mosaic technique of imitating compound eyesThis paper puts forward a registration method based on space coordinate. Aiming at multiple frames static scenes, a new image mosaic method imitating compound eyes is presented based on wavelet multi-scale analysis. A new ruler of image mosaic is founded. Aiming at multiple frames video sequences with moving objects, a new image mosaic method imitating compound eyes is presented based on image’s characteristic. Now. some mosaic algorithms are local optimizing. These two methods overcome thosedisadvantages, and are global optimizing.3. Method of infrared image processing imitating fly visionThe Figure-Ground Discrimination System is discussed; the formulas of output response are given about Large-Scene system and Small-Scene system. This paper discusses the method about fly lateral inhibitory networks applying to image preprocessing. Aiming at target detecting mechanism of fly vision, we propose a new method for infrared image segmentation using the fuzzy entropy. Its segmentation velocity is fast, and can be satisfied with the feature of real time.4. Infrared target recognition imitating fly visionThis paper proposes a new method based on edge moment invariants and the infrared target feature for feature extraction. A new target recognition method based on support vector machine is proposed. Studies show it is faster than the others, and it is proved to be effective.5. Applying fly vision system to imaging guidanceA new physical model of imaging seeker is proposed, this seeker imitates fly’s compound eyes, can realize 360?searching range. According to this model, we simplify the registration method of image. At last, a synthetical application flow is given about fly vision applying to imaging guidance. It expands the thought for latter research.更多还原