复杂背景下基于自适应熵流的红外弱目标检测与跟踪技术研究

Under the effect of solar variation、thermal radiation distribution and atmospheric attenuation, the grey value of interference sources is close to or equal to the target gray value; with the very long distance between the imaging system and the target, the target is dotted or mottled in the image; in addition, the target motion is often unknown and abrupt. All these factors cause the negative effect of dim target recognition. This project intends to put forward an algorithm based on adaptive entropy flow to detect and track dim target under complex background. The research is divided into three steps: Firstly, aiming at the effect of clutter on image filtering, the paper studies the image converting theory from brightness pattern to entropy pattern, then the weighted entropy model of spatial adjacency and direction priority is presented to preserve the topology structures of image and improve the target signal-to-clutter ratio; Secondly, aiming to the different regions where pixel values change drastically or smoothly, the paper researches the nonlinear smoothing and local constraint criterion of entropy flow, then the non-linear diffusion motion model of adaptive entropy flow is constructed to represent the motion status of target; Thirdly, according to the uncertainty and mobility, the paper studies the target correlation function, builds the sequential filtering model fusing movement and unmovement characteristics, and tracks the dim target through accumulation of target energy and trail search. In the civil and military field of target detection、recognition and tracking, intensive study on adaptive entropy flow can provide theoretical foundation for robustly detecting and tracking dim target.

受太阳光照变化、热辐射分布与大气衰减影响,干扰源的灰度值接近甚至等于目标灰度；当成像系统与目标相距较远时，在图像中呈现为点状或斑点状；此外，目标运动往往是未知和突变的，它们会产生弱目标识别的负面效应。本项目拟提出复杂背景下基于自适应熵流的弱目标检测与跟踪算法。研究分为三个方面：（1）针对杂波对图像滤波影响，研究图像从灰度模式转换为熵模式的理论，提出空域邻近与方向优先的加权熵模型，从而保持图像拓扑结构与提高目标信杂比；（2）针对不同区域灰度变化平滑或强烈的特点，研究熵流的非线性平滑与局部约束准则，构造自适应熵流的非线性扩散运动模型，描述目标运动状态；（3）针对目标的不确定性与机动性，研究目标关联函数，构建运动与非运动特征融合的序贯滤波模型，采用目标能量累积与轨迹搜索实现弱目标跟踪。 在民用与军事的目标探测、识别与跟踪领域，深入开展自适应熵流研究能够为鲁棒地检测与跟踪弱目标提供理论支撑与依据。

远距离红外成像过程受到自然环境干扰、成像系统性能、成像距离与目标随机运动等因素影响，使得获取的红外图像可能存在模糊、噪声、杂波和灰度变化剧烈等现象，导致了目标虚检与漏检的问题。首先，研究了保持拓扑结构的图像预处理技术，提出了空域邻近与方向优先的仿生视觉加权熵模型。图像灰度信息矩阵以加权熵信息矩阵来描述，加权因子刻画了图像空域灰度分布特征与区域显著性特性，实现了图像从灰度模式变换为熵模式，有效抑制了杂波；其次，研究了目标突发或平稳状态的熵流运动估计技术，提出了自适应熵流的目标运动估计模型。构建了数据项与平滑项的能量泛函，采用各向异性扩散方法平滑图像的局部区域，数据项的加权因子与图像局部区域平滑程度变化趋势相同，从而刻画了目标运动状态的逼近；最后，研究了类属目标时空特征趋近性的跟踪技术，提出了多特征融合的序贯滤波模型。构建了目标关联的特征评价函数，在邻近空域特征前提下，假如评价值适配，可能为目标, 再经目标轨迹能量累积的序贯滤波方法检测弱目标。.因此，随着红外目标探测距离的日益增大与分辨能力的不断提高，本项目的研究成果有利于增强小目标的稳健检测能力。