面向复杂场景低空慢速小目标检测方法研究

Detecting the low-altitude, slow-speed and small target under complex scenario has become a difficult and hot problem in the aspect of air defense security. Because low-altitude, slow-speed and small target has the characteristics of small size, weak signal, complex scenario, slow-moving, which increase the difficulty of target detection and precaution, the existing methods based on optical and radar equipment are no longer applicable and effective. For all this, considering the low-altitude, slow-speed and small target (such as small unmanned aircraft and aviation models) as study subjects, on base of infrared and visible image detection theory, by means of signal enhancement, background or clutter suppression and information fusion, combined with feature extraction technology, this project will focus on the following research: 1) reveal the separation mechanism of complex interference of cloud and haze, set up haze removal methods based on physical model and cloud suppression methods based on lateral inhibition; 2) clarify the sequential accumulation mechanism of target energy, and establish the sequential accumulation methods based on the spatiotemporal model and the superposition principle; 3) build the overall detection and local detection methods and reveal the twice determination collaborative mechanism between them, and give twice determination method for suspected small targets. This project aims to achieve some innovative research results, and lays a theoretical foundation for the low-altitude, slow-speed and small target detection methods in practical application.

复杂场景下低空慢速小目标检测已成为空防安保的难题和研究热点。由于低空慢速小目标具有体积小、信号弱、场景复杂、运动慢等特点，增加了该类目标侦测和防范的难度，使得现有无论是基于雷达探测设备还是光学探测设备的目标检测方法显得不能适用，效果不佳。为此，拟以低空慢速小目标（如小型无人机和航空模型）为对象，以红外图像检测和可见光图像检测理论为基础，以信号增强、背景或杂波抑制和信息融合方法为手段，结合特征提取技术，重点开展以下研究：1）研究云雾等复杂干扰的分离机制，建立基于物理模型的去雾方法及基于侧抑制机理的云层抑制方法；2）研究目标能量贯序积累机理，基于时空模型和叠加原理建立慢小约束下目标能量积累方法；3）研究整体检测和局部检测方法以及协同整体检测与局部检测的二次判定机制，并给出疑似小目标的二次判定方法。旨在取得一些创新性研究结果，为低空慢速目标检测方法在实际中的推广应用奠定理论基础。

由于低空慢速小目标具有体积小、信号弱、场景复杂、易操控、飞行高度低、运动慢等特点，成为低空侦测和防范的难点，其漏检与失控可能会导致严重的安全威胁与后果。本项目针对低空慢速小目标检测问题开展了深入研究，提出了：1）基于信息积累和近邻关联的目标检测方法；2）基于时域滤波和图关联的目标检测方法；3）基于目标重构和时空关联的目标检测方法。此外，还对目标分割和目标识别等问题开展了深入的研究，为目标检测提供了新思路，提出了：1）基于活动轮廓与学生t-分布混合模型的分割方法；2）基于多区域活动轮廓的分割方法；3）基于空间位置关系的目标识别上下文模型；4）基于小波变换与稀疏傅里叶变换的光场重构方法。研究成果为低空慢速小目标检测方法的推广应用，提供了重要的理论支撑和知识储备。