复杂背景下的红外弱小目标的检测

【作者】 曾雅琼；

【导师】 陈钱；

【作者基本信息】 南京理工大学 ， 光学工程， 2012， 硕士

【摘要】 作为红外搜索与跟踪领域很重要的一个课题,在复杂背景的红外图像中对小目标检测的一系列瓶颈问题和关键技术受到了国内外专家学者的高度重视。本文研究的即是这个领域的两项关键技术：背景抑制技术和基于多帧的弱小目标检测。本文首先提出两种改进的滤波算法。双边滤波,去噪同时也能较好地保留边缘,结合这一优点,本文对其滤波算子添加一个能保护小目标的模板；规整化的各向异性自适应滤波,对不同的结构具有自适应性,结合Robinson滤波的判据条件,二者互补。分别利用这两种算法针对不同的场景进行仿真,发现都有很好的鲁棒性。且与传统算法相比有较低的虚警率、较高的信杂比增益和背景抑制因子。在图像分割阶段,提出一个基于平均方差加权熵的自适应阈值,利用大量实验数据讨论参数的选取来得到最好的效果。将滤波后的图与原图像的特征相结合,与其他分割算法相比,能更好地去除虚假目标。接下来,结合Kalman滤波来对帧与帧之间的数据关联进行维持与更新,检测过程中,利用航迹评价因子进一步剔除虚假航迹,降低计算复杂程度,并最终得到真实目标的轨迹。最后给出整个算法流程的实验证明,发现整个算法确实能正确有效的检测出红外弱小目标的正确轨迹。更多还原

【Abstract】 As an important topic in infrared(IR) search and tracking, a series of bottleneck problems and key techniques of IR small target detection in complex background are highly valued by the experts at home and abroad. This dissertation concentrates just in the two key techniques:the suppression of background clutter and the detection of the dim small moving targets in multi-frames.At first, this paper proposed two improved filter algorithm. Bilateral filtering, de-noising and effectively keeping the edge, is added a template which can protect small target with its filter operator; Regularizing filtering(RegAF) is adaptive to different structures.A integrated and complementary of RegAF and Robinson filtering is proposed by this dissertation. According to different scene, the simulations both show good robustness. And compared with the traditional methods the two have lower false alarm rate, higher gain of Signal-to-Clutter Ratio and Background Suppression Factor(BSF).In the image segmentation stage, an adaptive threshold value based on mean variance weighted information entropy is proposed with its parameters obtained by a lot of experimental data. With combination of characteristics of original image and filtered image, it is better able to remove the false target compared other segmentation algorithms.Next, maintenance and update the data connection of sequential images via kalman filter. During detection, track evaluation factors further eliminate false track, reduce the complexity of the calculation, and finally the real goal of the track is obtained.The results of whole algorithm are given in the end, which show that the proposed algorithm can detect right track of infrared dim-small target validly.更多还原