基于红外图像信息融合的目标检测和识别技术研究

【作者】 李秋华；

【导师】 沈振康； 李吉成；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2002， 硕士

【摘要】 随着精确制导武器的发展，人们希望在小目标阶段，即目标与传感器相距较远时能检测和识别出目标，以进行有效的防御和攻击。而从红外图像传感器获得的图像经过预处理之后，得到的仅是一些点或斑点目标。此时如果只凭借于单一的图像传感器所获得的图像序列信息检测目标的可靠性往往不高，同时也很难准确地把拟攻击的真目标从各种虚假目标和随机干扰中区别出来。而小目标的高可靠性检测和识别对后续的目标捕获和跟踪是极其关键的。基于上述认识，本文对低信噪比小目标的高精度检测和识别进行了研究，提出了基于多传感器信息融合的小目标检测和识别算法。该算法的具体思路为：首先对来自两个不同波段(波长分别为3～5μm，8～12μm)的红外图像序列进行双模融合检测和跟踪，获得目标的速度信息和灰度信息；然后采用多传感器信息融合理论对小目标的双模图像信息进行了融合识别和分类处理，以识别出拟攻击的真目标。 论文的研究工作可概括如下： 第一章介绍了本课题的应用背景、多传感器信息融合理论及该理论在目标检测和识别中的应用；然后给出了几种传统的目标检测和融合识别方法。 第二章在详细讨论了各种单帧和多帧目标检测技术的基础上，提出了双波段融合检测方法，并对该算法的检测效果进行了理论分析和试验比较。 第三章提出了一种基于D-S证据理论的多传感器时间—空间信息融合识别算法，并采用该方法对实际的双波段图像进行目标融合识别仿真计算，较大地提高了强干扰条件下的红外小目标识效率和可靠性。 第四章在讨论模糊综合理论及其在多传感器决策级信息融合中的应用的基础上，提出了一种基于模糊综合的红外图像目标融合识别方法，然后采用该算法对双波段图像序列进行目标识别仿真试验，并给出了该算法的识别效率和改进方法。 在本文最后，对基于多红外双色图像信息融合的红外小目标检测和识别算法作了总结，并简单分析了融合检测和识别算法的实用性和应用前景。更多还原

【Abstract】 As the development of the modern remote distance tacking weapon, we hope to recognize target in the stage of small target, that is the stage in which the distance between target and sensor is very far, so that to defend and tack effectively. But we can only get the information of kinematics and gray of the dot targets by pre-processing the image serials from IR sensors in the stage of target detection .It is very difficult to distinguish the real target from all kinds of virtual target and stochastic clutter if we use only the information from single sensor. But the precise recognition in the stage of small target is very important for the following target capturing and target tracking. The paper has done some works as the follows on the problem: As for the problem that how to improve the precision of IR small targets detection and the probability of recognition, has studied the imaging characteristic of IR dual band imaging system and presented the algorithm of detection and recognition for IR small targets on the base of multi-sensor information fusion.The whole algorithm’s idea consists of two aspects: the first, processed the IR image sequences from the two different IR bands by detecting on the fusion of dual band and precise tracking to get the information of velocity and grayscale of targets; the second, fusing the information of the small targets form IR dual band image using the theory for multi-sensor information fusion to identify the real target to attack at the high probability.In the first chapter, a general introduction is given on the background for applying of the project and the theory for multi-sensor information fusion and the application of the theory to detection and recognition of IR small targets. Then given a concise presentation for several kinds of target detection and fusion recognition.In the second chapter, a detailed discussion is presented here for the IR small detecting technique based on the data fusion of dual band IR image information, then to preprocess images using the algorithm. The results are satisfying.In the third chapter, the method of IR small target based multi-sensor temporal-spatial information fusion(applying D-S evidence combination theory). And fused the real image sequences from the IR dual band imaging sensors using the method, improved the probability and performance of IR small target under strong clutter greatly.In the fourth chapter,presented a method based on fuzzy combination IR image target fusion recognition ,then applied the method to the image sequences from the IR dual band imaging system for target recognition, and discussed the recognition performance and the improving ways for the method.In the end of the paper, summarized the algorithms of IR small target detection and recognition based更多还原