【作者】 张峰；

【导师】 刘上乾；

【作者基本信息】 西安电子科技大学 ， 物理电子学， 2010， 博士

【摘要】 随着红外成像自动目标识别(ATR)技术在精确武器系统上的广泛应用,世界各国都在加速发展红外成像ATR技术的研究和装备的研制。本文根据“十一五”国防重点预研项目,对红外成像ATR系统研制过程中涉及到的若干问题作了较为深入的研究,主要研究成果如下：针对红外焦平面阵列器件(IRFPA)在大动态范围内应用的问题,提出了基于非线性快速卡尔曼滤波的非均匀性校正算法,该算法能对IRFPA实现非均匀性和非线性的双重校正,且运算量较小。为满足系统在某些实际场合应用的需要,针对传统基于场景统计的非均匀性校正算法的不足提出了一种基于平稳小波变换的非均匀性校正算法,该算法能在场景变化不充分的条件下对IRFPA进行非均匀性校正。由于红外图像在传输过程中易受到噪声的污染,加之红外成像本身具有图像细节模糊不清的特点,严重影响了后续目标检测以及目标匹配的精度,为此提出了一种基于联合直方图均衡及图像融合的红外图像增强算法,该算法能够有效抑制噪声,提高图像的对比度；另外该算法通过对原始图像进行非线性外推处理得到新的细节成分,在增强对比度的同时增强了图像目标的细节。针对传统帧累加以提高图像信噪比的方法易造成图像中运动目标模糊的情况,提出了一种改进的基于图像序列的图像增强算法,该算法通过准确的提取图像中目标的运动场,通过带有运动补偿的时空域滤波完成红外图像的2D-TDI增强,仿真实验的结果验证了算法的有效性。针对复杂背景下低信噪比红外弱小目标检测问题,提出了基于各向异性判决和双边滤波的红外弱小目标检测算法。该算法能对图像背景中精细部分进行有效预测,取得了较好的背景抑制效果。针对红外成像ATR系统需要精确的目标轮廓的要求,提出了一种基于蚁群算法的快速二维模糊熵图像分割算法,该算法使用二维模糊熵的设计思想并将其改进,推导出快速算法,最后使用蚁群算法优化其阈值求解,达到了快速、准确提取目标轮廓的目的。由于红外小目标的检测易受到各种虚假目标及随机干扰的影响,为此提出了一种基于模糊D-S证据合成理论的双色红外小目标识别算法,使用双波段红外成像,克服单波段红外成像易受干扰的缺点,并将模糊集理论与D-S证据合成理论融合,提高了对获得的目标信息识别的能力。针对红外面目标识别的问题,着重研究了小波矩不变量在红外成像目标识别方面的应用及其性能,该算法能对位置、尺度和视角发生变化的目标进行识别,仿真实验结果证明了该算法的有效性,取得了较满意的效果。此外,针对复杂战场应用环境,设计并研制了一套高帧频的红外成像ATR实时信号处理系统,该系统能实时稳定的完成红外成像ATR系统所需的一系列算法,具有工作帧频高、信号动态范围大、实时性强、处理精度高和灵活性等优点。更多还原

【Abstract】 Along with the widespread application of infrared imaging automatic target recognition(IR ATR) on precision weapon system, the technology of IR ATR and its equipments are being rapidly developed all over the world. To meet the demand of the national defense key advanced research project for military, some problems regarding the IR ATR system are deeply studied in this dissertation, and some valuable results are obtained which are summarized as follows:In view of the application of infrared focal-plane array device (IRFPA) in a large dynamic range, a noniniformity algorithm based on no linear fast Kalman filtering is proposed which can implement both nonuniformity and nonlinearity correction to IRFPA with less calculation. Aiming at the applications in some special environment of the inadequate scene change as well as the meet of system practical application, a correction algorithm based on stationary wavelet transform is presented which can be carried on under such circumstance.The infrared imagery is easy contaminated by noise in the transmission process, in addition the infrared imagery has the properties of blurring and Details unclear, which affect the accuracy of target detection and matching seriously. For this reason a new adaptive enhancement algorithm is proposed based on joint histogram equalization with image fusion, which can suppress noise and enhances the contrast effectively. Moreover new details is added by the use of nonlinear filter which enhances the detail while enhances the contrast. Aiming at the shortage of motion target edge blur caused by traditional frame accumulation SNR enhancement method, a new image enhancement algorithm is proposed based on image sequence. This algorithm can extract the motion field accurately of the target and then time-spatial filtering is applied to implement the image enhancement process. The validity of the method is validated by experimental results.Facing with difficulties of the infrared dim target detection in the condition of complicated background and low image SNR, a detection algorithm based on anisotropic judgment and bilateral filtering is presented. This algorithm can make an effective prediction of image background with subtle differences, and achieves a good suppression result. In view of the accurate Object Figure which IR ATR system needed, a fast two-dimension fuzzy entropy image segmentation algorithm based on ant colony algorithm is proposed. The formula of the fast two-dimension fuzzy entropy is derived, and then ant colony algorithm is employed to optimize the solving of the image threshold so that the object figure is extracted quickly and accurately.As the detection of infrared small target can be easily affected by various false target and random disturbance, a dual band infrared small target recognition algorithm is proposed based on fuzzy Dempster-Shafter evidence theory. The use of dual band infrared imaging overcome the shortage of single band imaging, and the combination of fuzzy sets theory and Dempster-Shafter evidence theory improves the recognition capability on obtained target information. Aiming at large target recognition, the application of wavelet moment invariant in infrared target recognition and its performance are deeply studied in this dissertation. This algorithm is able to recognize target with variations in size, position and orientation effectively. Experimental results show the superiority of the algorithm, and achieve satisfactory results.In addition, considering the application in complex battlefield environment, a high frame-rate IR ATR real-time signal processing system is also proposed, which can implement a series of algorithms that IR ATR system needed. This system has the advantage of high frame-rate, large dynamic range, high operation speed and precision, vigorous flexibility and versatility, etc.更多还原