基于低秩引导滤波和空时显著信息融合的海空背景红外小目标检测方法研究

This project studies the target detection algorithm for high-speed and high-maneuvering small target to provide technical support for infrared early warning and reconnaissance systems. Improving the robustness and accuracy of the target detection algorithm is the core problem of the existing research. The main contents of this project are small infrared target detection method based on background prediction and saliency analysis, which can increase the performance of the existing detection method and provide new ideas for infrared high maneuvering small target detection. We will study three main problems: 1. The novel background prediction model based on low rank sparse decomposition and tuning guide filtered is proposed to effectively suppress background clutter and increase contrast of the target and SNR(Signal noise ratio); 2. To fully tap the motion saliency information of the infrared images, the saliency analysis method based on motion vector and human visual mechanism is proposed in temporal domain; 3. To achieve precise positioning of the target region, fusion rules based on saliency dynamic competition for spatial and temporal information are proposed, and complex saliency analysis model is built. Solve the above three problems, we will be able to effectively improve the robustness and accuracy of infrared small target detection method. The related outcomes will improve the infrared target detection and tracking capability of early warning and reconnaissance systems, and assist the decision-making bodies to analyze and judge the emergency situations.

本项目以高速、高机动小目标为研究对象，旨在研究相应的目标检测算法，用于为红外预警侦察系统提供高效的技术支持。如何提高目标检测方法的鲁棒性与精度，是现有研究面临的核心问题。本项目的主要内容是研究基于背景预测和显著性分析的红外小目标检测方法，为改善现有检测算法性能，提高算法效率提供新的思路。为此将研究三个主要问题：1、研究基于低秩稀疏分解和调谐引导滤波的背景预测模型，有效抑制复杂背景杂波影响、提高目标对比度和信噪比；2、研究基于运动矢量分析的时域显著性分析方法，利用人眼视觉机制，充分挖掘源图像的运动显著性信息；3、研究基于动态竞争的空域和时域显著信息融合规则，构建复合显著性分析模型，实现目标区域的精确定位。解决了这三个问题，将能有效提高红外小目标检测方法的鲁棒性和精度，课题相关成果将提高红外预警侦察系统的目标检测跟踪能力，辅助决策机关分析判断敌情。

红外小目标图像背景复杂、受杂波干扰严重、目标检测难度高。针对上述问题，本项目的主要工作及创新如下：（1）构建基于低秩引导滤波的背景抑制模型，利用改进的鲁棒性主成分分析算法对图像降维分解，得到低秩背景块和稀疏前景块，实现背景与目标分离；在此基础上，提出基于引导滤波和模糊算法的红外背景抑制算法，兼顾了去噪和抑制背景，弥补了传统空间域背景抑制方法的缺陷；为了在杂波中甄别目标、避免漏检，提出基于密度峰值搜索和局部镶嵌梯度漫步的红外小目标检测算法。（2）针对红外图像目标边缘模糊、缺少颜色等特征，导致传统目标检测方法检测精度低的问题，面向红外弱小目标的运动矢量建模。采用Space-time saliency方法提取红外目标的运动状态数据，将局部显著性分析应用于运动小目标检测。（3）针对非结构化场景中含有大量非确定的状态，导致基于显著性分析的目标检测定位边界的精度低，算法普适性受限的问题，提出时空显著性融合分析模型。将显著目标检测问题转换为像素块的显著值优化问题。提出基于区域概率统计和显著性分析的目标检测，降低了虚警率，提高了检测精度。实验证明，本项目提出的检测算法有效提高了红外小目标检测的鲁棒性和精度，课题相关成果为提高红外预警侦察系统的目标检测跟踪能力提供了理论支持。