基于BEMD和时空融合的红外弱小目标检测算法研究

发布时间：2018-02-13 09:38

本文关键词： 红外弱小目标检测 BEMD 移相差分算法时空融合分割　出处：《华中科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：红外探测系统是远程打击系统和预警系统的重要组成,但是复杂背景下低信噪比红外弱小目标的检测一直是一项非常困难的课题,正因为如此,该课题成为远程精确打击系统和早期预警系统的核心研究内容之一。针对这个难题,国内外很多学者提出了很多富有成效的红外弱小目标检测算法,但是现阶段这些算法也都存在一定的局限性。本文提出了一种基于BEMD和时空融合的复杂背景低信噪比弱小目标检测新方法,其中EMD算法是一种分析一维非平稳信号的有力工具,不同于傅里叶变换、小波变换等数据分析方法,EMD是完全的数据驱动的自分解过程,不需要人为构造变换的核函数,具有很强的自适应性。本文将一维EMD算法推广到二维应用中,利用BEMD处理红外弱小目标图像,得到含有小目标特征信号的分解结果,再和本文提出的移相差分算法进行融合,最后对融合结果进行分割,检测出红外图像中的弱小目标。传统BEMD存在不一致收敛和模态混叠问题,这两个问题极大的限制了BEMD的应用。针对这两个难题,本文借鉴项目管理中常用的“鱼骨图分析方法”,找出极值点提取不足是造成BEMD过程中迭代条件振荡,算法不一致收敛以及模态混叠问题的主要原因,并针对传统BEMD的不足,提出改进的BEMD算法,通过对比试验,本文改进的BEMD较传统BEMD有着更快的收敛速度和分解效率,有效的克服了不一致收敛和模态混叠问题,为BEMD应用于红外弱小目标检测奠定了基础。针对现有红外弱小目标图像背景抑制算法的不足,本文开创性的提出移相差分算法,该方法基于单帧图像处理,采用移相的方法制造出小目标“运动”的性质,再将不同相位差值信号融合,可以极大的抑制红外背景,是一种简单高效的背景抑制新算法。针对小目标特征信号及背景抑制图像中存在少量杂波的问题,本文提出的基于BEMD和时空融合方法滤去背景杂波,可以极大地提高分割效果。
[Abstract]:Infrared detection system is an important component of long-range strike system and early warning system, but the detection of infrared small and weak targets with low signal-to-noise ratio (SNR) in complex background is always a very difficult task. This subject has become one of the core research contents of remote precision strike system and early warning system. In view of this problem, many scholars at home and abroad have put forward many effective infrared dim target detection algorithms. However, these algorithms also have some limitations at present. In this paper, a new method based on BEMD and spatio-temporal fusion is proposed to detect small and weak targets with complex background and low signal-to-noise ratio (SNR). The EMD algorithm is a powerful tool for analyzing one-dimensional non-stationary signals. Unlike Fourier transform, wavelet transform and other data analysis methods are completely data-driven self-decomposition processes, which do not require artificial construction of the kernel function of transformation, and have strong self-adaptability. In this paper, one-dimensional EMD algorithm is extended to two-dimensional applications. The infrared dim target image is processed by BEMD, and the decomposition result with small target characteristic signal is obtained, and then fused with the phase shift difference algorithm proposed in this paper. Finally, the fusion result is segmented. The weak targets in infrared images are detected. There are inconsistent convergence and modal aliasing problems in traditional BEMD, which greatly limit the application of BEMD. This paper draws lessons from the "fish bone map analysis method" commonly used in project management, finds out that the shortage of extreme point extraction is the main cause of iterative condition oscillation, algorithm convergence and modal aliasing in the BEMD process, and aims at the deficiency of traditional BEMD. An improved BEMD algorithm is proposed. The improved BEMD has faster convergence speed and decomposition efficiency than the traditional BEMD, and effectively overcomes the inconsistent convergence and modal aliasing problems. This paper lays a foundation for the application of BEMD in infrared dim target detection. In view of the shortcomings of existing background suppression algorithms for infrared dim and small target images, this paper proposes a novel phase shift differential algorithm, which is based on single frame image processing. The method of phase shift is used to create the "motion" property of small target, and then the signal of different phase difference can be fused, which can greatly suppress the infrared background. It is a simple and efficient new algorithm for background suppression. Aiming at the problem of a small number of clutter in the feature signal of small target and the image of background suppression, this paper proposes a method based on BEMD and spatiotemporal fusion to filter background clutter, which can greatly improve the segmentation effect.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41

【参考文献】