基于红外搜索跟踪系统的低慢小目标检测技术研究
发布时间：2018-12-17 18:16
【摘要】：红外搜索跟踪IRST系统因为其在雷达探测的低空盲区依然能够对低慢小目标进行检测,因此被广泛的用作雷达的补盲设备。而低慢小目标往往是具有较大威胁能力的目标,当前对低慢小目标的检测不断提出更高的要求,因此IRST的低慢小目标检测能力是评估其性能的重要指标,本文结合IRST工作的实际,提出了切实可行的一套目标检测算法。本文第二章讨论了提取天空背景的工作,针对IRST捕获图像的复杂性和实际情况,设计了一种全天空场景、全地面场景以及天空地面混合场景的识别算法,通过这样的算法可以提取出全天空背景图片和天空地面混合背景的图片,又进一步研究了天空地面混合场景中天空背景区域的提取方法,由于传统分割算法的局限性,本文设计了Gabor滤波描述局部场景的方法提取天空地面混合场景中的天空背景区域。通过对比发现本文提出的天空区域提取方法具有可靠性高实用性强的特点。本文第三章讨论了单帧目标的检测算法,本文提出了一种基于多尺度目标模型的目标检测算法,同时通过实验对比发现本文算法在分割门限取的最低的时候引入的虚警点数量不是太多,同时对天空背景中的干扰面目标和残留的地面背景具有一定的抑制作用。另外,本文算法检测出来的目标保持了弱小目标的成像特点,在图像上呈现孤立点分布,为多帧联合检测提取真实目标算法提供了支持。本文第四章研究了多潜在目标条件下提取真实目标的工作,针对真实目标在一定时间内多帧关联图像中呈现匀速直线运动特点,而噪声则在多帧图像中呈现随机分布的特点,本文提出多步PPU滤波算法和多步LPPU算法对潜在目标点进行监测,从中提取出符合匀速直线运动的观测点作为真实目标,排除杂乱分布的噪声虚警点,通过实验发现该算法的检测效果比较好,在杂波强度较高的时候依然能够具有较好的检测正确率。
[Abstract]:Infrared search and tracking (IRST) system is widely used as a blind equipment for radar because it can detect low slow small targets in the low altitude blind area detected by radar. However, low slow small target is often the target with great threat ability, so the detection ability of low slow small target in IRST is an important index to evaluate its performance. Based on the practice of IRST, a set of feasible target detection algorithm is proposed in this paper. In the second chapter, the work of extracting sky background is discussed. Aiming at the complexity and actual situation of IRST capture image, a recognition algorithm of whole sky scene, all ground scene and sky ground mixed scene is designed. Through this algorithm, we can extract the whole sky background image and the sky ground mixed background image, and further study the sky background region extraction method in the sky ground mixed scene, because of the limitation of the traditional segmentation algorithm. In this paper, Gabor filter is designed to describe the local scene. By comparison, it is found that the sky region extraction method proposed in this paper has the characteristics of high reliability and high practicability. In the third chapter, we discuss the single frame target detection algorithm, and propose a multi-scale target model based target detection algorithm. At the same time, the experimental results show that the number of false alarm points introduced in this algorithm is not too many when the threshold is the lowest, and it can restrain the interference surface target and the residual ground background in the sky background to a certain extent. In addition, the target detected in this paper keeps the imaging characteristics of small and weak target, and presents outlier distribution on the image, which provides support for multi-frame joint detection and extraction of real target algorithm. In the fourth chapter, we study the work of extracting real targets under the condition of multiple potential targets, aiming at the characteristics of the real targets showing uniform linear motion in multiple frames of correlation images in a certain time, while the noise is randomly distributed in multiple frames of images. In this paper, a multi-step PPU filtering algorithm and a multi-step LPPU algorithm are proposed to monitor the potential target points, from which the observation points which accord with the uniform linear motion are extracted as the real targets, and the noise false alarm points of the clutter distribution are eliminated. The experimental results show that the algorithm has a good detection effect and can still have a good detection accuracy when the clutter intensity is high.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41;TN219
[Abstract]:Infrared search and tracking (IRST) system is widely used as a blind equipment for radar because it can detect low slow small targets in the low altitude blind area detected by radar. However, low slow small target is often the target with great threat ability, so the detection ability of low slow small target in IRST is an important index to evaluate its performance. Based on the practice of IRST, a set of feasible target detection algorithm is proposed in this paper. In the second chapter, the work of extracting sky background is discussed. Aiming at the complexity and actual situation of IRST capture image, a recognition algorithm of whole sky scene, all ground scene and sky ground mixed scene is designed. Through this algorithm, we can extract the whole sky background image and the sky ground mixed background image, and further study the sky background region extraction method in the sky ground mixed scene, because of the limitation of the traditional segmentation algorithm. In this paper, Gabor filter is designed to describe the local scene. By comparison, it is found that the sky region extraction method proposed in this paper has the characteristics of high reliability and high practicability. In the third chapter, we discuss the single frame target detection algorithm, and propose a multi-scale target model based target detection algorithm. At the same time, the experimental results show that the number of false alarm points introduced in this algorithm is not too many when the threshold is the lowest, and it can restrain the interference surface target and the residual ground background in the sky background to a certain extent. In addition, the target detected in this paper keeps the imaging characteristics of small and weak target, and presents outlier distribution on the image, which provides support for multi-frame joint detection and extraction of real target algorithm. In the fourth chapter, we study the work of extracting real targets under the condition of multiple potential targets, aiming at the characteristics of the real targets showing uniform linear motion in multiple frames of correlation images in a certain time, while the noise is randomly distributed in multiple frames of images. In this paper, a multi-step PPU filtering algorithm and a multi-step LPPU algorithm are proposed to monitor the potential target points, from which the observation points which accord with the uniform linear motion are extracted as the real targets, and the noise false alarm points of the clutter distribution are eliminated. The experimental results show that the algorithm has a good detection effect and can still have a good detection accuracy when the clutter intensity is high.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41;TN219
【相似文献】
相关期刊论文 前10条
1 周冰;王永仲;孙立辉;何永强;;图像局部熵用于小目标检测研究[J];光子学报;2008年02期
2 任蕾;施朝健;冉鑫;;应用改进频率调谐的海上小目标检测方法[J];中国图象图形学报;2012年03期
3 林建|，
本文编号：2384614
本文链接：https://www.wllwen.com/kejilunwen/dianzigongchenglunwen/2384614.html
