基于四维空间多光谱信息融合的多个弱小运动目标检测与识别

Infrared dim moving targets with low SNR only cover several pixels. Target signal is fluctuating, and the texture, size and structure of the target are poor. These issues make it difficult to extract the effective features for target recognition. The targets are always immersed by clutter and noise in non-stationary and complicated background. The appearance and disappearance of the targets are also random. These factors make the target detection and recognition more difficult. The above problems become more intractable with the development of new electronic warfare and stealth technology. The four-dimensional space is proposed to process the infrared image. Two-dimensional location information, one-dimensional temporal information and one-dimensional spectral information are fused to accomplish the detection and recognition of multiple fluctuating dim moving targets in infrared multispectral image sequence. The multispectral images are firstly fused using dimension reduction algorithm, based on which a track-before-detect algorithm for multiple targets is proposed. The method of extracting robust features and the probabilistic neural networks-based technique for real-time recognition of multiple targets are studied. Real-time recognition of multiple airborne infrared fluctuating dim moving targets with low SNR are implemented. This achievement will be widely used in the military and civil fields including infrared precision-guidance, aviation detection, and infrared early warning, etc.

低信噪比条件下红外弱小运动目标成像面积很小，目标信号呈现出闪烁特性，缺乏纹理，尺寸和结构等信息，用于目标识别的有效特征很难提取，且在非平稳复杂背景条件下，目标常常被杂波和噪声淹没，目标的出现和消亡表现出随机性，这些因素增加了检测与识别的困难。随着新型电子对抗技术及隐身技术的发展，上述问题更为突出。本课题突破传统的解决方式，提出构建四维空间处理红外图像思想，通过融合图像的二维空间信息，一维时间信息和一维光谱信息，基于红外多光谱图像序列实现对多个闪烁弱小运动目标的融合检测与识别。本项目首先设计多光谱数据降维算法对多光谱数据进行融合，在此基础上提出一种检测前跟踪算法实现多目标检测，并对红外弱小目标的稳健特征提取技术及基于概率神经网络的多目标实时识别技术展开研究，最终实现低信噪比条件下空中多个红外闪烁弱小运动目标的实时识别。研究成果将在红外精确制导，航空探察，红外预警等军事和民用领域获得广泛应用。

本项目以多个运动弱小目标的识别问题为研究背景，提出基于时\空\谱多维特征模型的检测识别算法，解决传统算法在低信噪比条件下难于检测红外弱小目标的问题，具有重要的理论意义和实用价值。本项目主要研究内容包括：弱小目标及背景的多光谱特征模型和图像序列数据库的建立，多个闪烁弱小运动目标的多光谱融合检测算法研究，空中弱小目标的多光谱融合识别算法研究，以及算法的性能评估技术及验证平台的设计工作。项目提出了基于离散相模型的红外目标光谱特征建模方法和基于变阻力特性的目标运动特征建模方法，构建了目标的多维特征模型；并以此为基础，提出基于二维经验模态分解和贝叶斯滤波的多个弱小目标检测算法，有效解决了因目标闪烁、背景噪声等问题而造成的目标检测困难。同时，提出了基于概率神经网络的弱小目标识别算法，利用机器学习方法实现弱小目标的识别和跟踪。最后，设计了基于嵌入式处理器的弱小目标检测识别算法验证平台，实现了算法的高效、实时运行，为算法性能验证和工程应用提供了可行的解决方案。.项目研究达到了预期目标，取得了具有创新性的研究成果，并发表学术论文13篇，其中SCI检索8篇，EI检索4篇。为多个运动弱小目标识别问题的解决提供了理论基础和技术支撑。