基于视觉显著机制的复杂背景下红外弱小目标检测研究

The detection of infrared (IR) small target under complex background is usually interfered by the highlight background, complex background edges and random pointy noises. Existing detection algorithms are usually difficult to suppress various interference while enhancing the target, resulting in a low detection rate, high false alarm rate and poor real-time performance. To solve this problem, this project proposes a novel detection method based on human visual system (HVS), starting from the following two aspects: the definition of the new spatial local contrast and the auxiliary weighting function. First, the idea of background estimation is introduced to improve the accuracy of the reference when calculating the local contrast, and, with the background estimation, the novel spatial local contrast definition will be proposed to enhance real targets while suppressing complex backgrounds, so the detection rate can be improved and the false alarm rate can be suppressed, besides, the detection speed will also be improved by the row/column parallel filtering. Then, the local complexity in single frame and the time domain local contrast between multi frames will be utilized as an auxiliary weighting function to improve the detection performance further. The method proposed in this project has an advantage of simple principle, easy realization and adaptive to the characteristics of the small dim targets, and is helpful to realize the detection of infrared small target with high detection rate, low false alarm rate and real-time performance under complex background, so it has certain significance and value in the field of infrared precise guidance, early warning, etc.

在复杂背景下检测红外弱小目标时常常受到高亮背景、复杂边缘、随机点噪等因素的干扰，现有算法通常难以在增强目标的同时抑制各种干扰因素，存在检测率低、虚警率高、实时性差等缺陷。本项目针对这一难题，拟提出一种基于视觉显著机制的新型检测方法，从新型空域局部对比度定义和辅助加权函数两方面分别展开研究：首先，项目拟引入背景估计思想，提高对比度计算时参照选取的准确性，并结合估计得到的背景，提出能够同时抑制背景、增强目标的新型空域对比度定义，以提高检测率、降低虚警率，同时通过提出行/列并行滤波的思想，提高检测实时性；然后，项目还拟综合利用单帧内的局部复杂度和多帧间的时域对比度等信息对求得的空域对比度进行辅助加权，进一步提高检测性能。本项目所提方法具有原理简洁易实现、与弱小目标特征契合度高等特点，有助于实现复杂背景下高检测率、低虚警率的实时目标检测，在红外制导和预警等领域具有一定的意义和价值。

复杂背景下的红外弱小目标检测一直是红外图像处理领域的研究热点与难点，高亮背景、复杂边缘、高强度点噪声等因素很容易给目标检测带来干扰，导致检测率低、虚警率高，影响实际应用。本项目针对复杂背景下的红外弱小目标检测难题，从提高背景估计的准确度、改善参照本底的选取原则、优化对比度计算公式、引入更多加权函数等方面展开研究，致力于提高算法的检测率、降低虚警率，从而提高算法的检测性能，同时也在降低计算复杂度、提高算法检测效率等方面做了一些初步的探索。由于红外弱小目标检测在制导、预警、监控、监视等军事领域具有比较突出的应用价值，因此本项目的研究成果在国防和国家安全领域也具有一定的潜在价值，特别是项目所提出的三层滤波窗口应用、最接近值选取原则、比差联合对比度计算方法等重要思想，在提高检测性能、降低检测时间等方面具有比较明显的实际效果，并得到了众多研究者们的认可，所发表的学术论文具有较高的引用量。在后续研究中，应重点关注局部对比度算法与神经网络、深度学习等前沿方法的结合，充分使用局部信息、非局部信息、时域信息等联合对红外弱小目标进行检测，最大限度地提高复杂背景下的检测性能。