【作者】 李红艳；

【导师】 吴成柯；

【作者基本信息】 西安电子科技大学 ， 信号与信息处理， 2000， 博士

【摘要】 十几年来，图像低信噪比小目标的检测问题一直是光学和红外图像领域的研究热点。光学传感器和红外为被动传感器，在现代战争中具有较强的生存能力，但其作用距离较短。研究低信噪比图像中小目标的实时检测及跟踪算法，可实现扩展它们作用距离的目的。低信噪比小目标检测的困难在于：目标信噪比低，又无纹理特征可以利用，只能采用序列图像的处理方法，但已有算法的缺陷在于运算量大，不易于实时实现。本文致力于研究低信噪比小（点）目标检测与跟踪的新方法。 在分析了小目标检测难点的基础上，指出不仅可以利用多帧图像沿目标航迹积累能量，还可以沿目标的观测区域积累能量，以提高目标的信噪比。但由于目标出现的时刻、目标的位置、目标的大小和目标的运动速度均未知，导致已有算法的运算量均较大。 给出了基于小波的小目标检测方法。充分利用目标的信息，实现了在目标的观测区域内积累目标信号能量的目的。理论分析与仿真结果表明，该方法可有效地提高小目标（信噪比＜2）的信噪比。 提出了基于遗传算法的点目标检测方法。图像中低信噪比点目标检测与跟踪，从理论上讲是一个难题。因无形状信息可利用，无法采用传统的图像处理方法，只能沿目标航迹进行能量积累，来检测目标。由于目标的位置和速度均未知，需构造目标的候选航迹，从中寻找目标航迹，是一个搜索算法。本文将遗传算法引入点目标的检测；设计了适于点目标检测的编码方案；依据点目标检测的特点，设计了相应的交叉方法、变异方法、适应度函数；在大量仿真的基础上，给出交叉概率、变异概率、群体规模等。仿真结果表明，该算法运算量小，可有效地检测信噪比低于2的点目标航迹。 提出了一种新的遗传算子—合格个体保留法，可以避免因目标信噪比过低而导致目标航迹的漏检，有利于遗传算法的收敛。 提出了用于图像低估噪比小目标跟踪的算法，即基于α—β滤波与多假设检验相结合的多目标跟踪算法。仿真结果表明，该算法可有效地同时检测并跟踪信噪比低于2的多条航迹。 最后提出了遗传算法与截断序贯似然比检验相结合的点目标检测与跟踪算法以及小波变换与遗传算法相结合的小目标检测算法。更多还原

【Abstract】 The detection and tracking of dim small targets in the optical and infrared images has been the subject of intense investigation for not more then two decades. The optical and infrared sensors are passive sensors, which are valued for their strong survival capacity in battlefields, but their maximum detection range is critical. The basic problem inherent to extent the detection range is the detection of small, low observable, moving targets in images and subsequent estimation of the target trajectories. The small spatial extent of these targets limits the information content of targets signature precluding the use of traditional pattern matching approaches, and the signal-to-noise ratio is sufficiently low that detection specifications cannot be met by an analysis of a single image frame. Most of approaches presented recently need high computation power are not Suit for real time implement. This dissertation address the problem of designing new efficient and effective image sequence processing schemes that will successfully detect and track small (point) targets with very low signal-to-noise ratio (SNR). After analyzing the difficulties of detection dim small targets, it points out that the small target signal energy should be integrated along its trajectory or in its extent. However, the position, size and velocity of an object in an image are initially unknown. These result in high computational requirements. An effective algorithm to detect dim small targets in images based on wavelet transform is presented here. The extent information is well be used to integrate the target抯 signal energy. Theoretical analysis and simulation results show that SNR can be successfully increased. A new dim point target detection algorithm based on Genetic algorithms (GAs) is proposed here. To a point target there is no extent information can be used. It is only can be detected by integrating signal energy along its path. How to find the objet path is a search problem. GAs is also a search problem. We induct GAs to the difficult problem of point targets detection. The code scheme and the operation of crossover, mutation and selection suitable for dim point targets detection are designed here. The simulation results show that it can detect and track targets computational efficiently with SNR<2. Eligible individual strings retained as a new genetic operation is proposed here. It avoids target tracks with very low SNR missing. It is in favor of GAs convergence. An algorithm, is also proposed for tracking low observable small (point) multi-targets. It combines multistage hypothesis tracking and intensity filtering to track moving multi- targets with SNR smaller than 2. At last, an algorithm, which combines Gas and truncated sequential probability ratio test, is presented here to detect and track point targets with low SNR. Another algorithm based on Wavelet transforms and Gas is proposed for dim small target detection.更多还原