SIA OpenIR  > 光电信息技术研究室
复杂背景下目标检测与抗干扰跟踪方法研究
Alternative TitleResearch on target detection and anti-jamming tracking method under complex background
张祥越
Department光电信息技术研究室
Thesis Advisor丁庆海
Keyword红外末制导 红外小目标检测 抗干扰目标跟踪 相关滤波 卷积神经网络
Pages132页
Degree Discipline模式识别与智能系统
Degree Name博士
2020-05-19
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract为了提高复杂背景下的目标检测与跟踪能力,本文以空空红外成像末制导为研究背景,开展复杂背景下目标检测与抗干扰跟踪技术研究,本文的主要贡献如下:1. 针对复杂背景下小目标检测能力不足的问题,提出了一种基于一阶方向过零点检测的红外弱小目标检测方法。该方法基于红外探测器在远距离的成像特性,对小目标和背景杂波的成像差异进行了分析,设计了一种一阶方向导数滤波器,该滤波器在多种不同方向对图像进行滤波,提高了小目标处的信号强度;随后利用此滤波器并基于过零点检测的思想来检测小目标。实验结果表明,该方法可以有效提高图像的信杂比并抑制背景杂波,能够准确检测出复杂背景下的红外弱小目标。2. 提出了一种基于三阶张量RPCA的红外弱小目标检测方法。为了保护像素之间的相关性,该方法首先提出了一种张量块模型将数据从二维空间转移到三维的张量空间。随后利用小目标图像的稀疏性与背景杂波的低秩性将小目标检测问题转化为数值优化问题,最后采用RPCA技术分离出小目标图像与背景图像。实验结果表明,该方法可以有效地从原始图像中提取出小目标,尤其在信号强度弱、小目标特征不明显的场景中表现出良好的小目标检测能力。3. 针对末制导中目标尺度不断变化的特点,从实际工程应用出发,开展多尺度点目标检测和多尺度面目标检测方法研究。在多尺度点目标检测方面,提出了一种基于中心域与邻域局部对比度的多尺度点目标检测方法。在多尺度面目标检测方面,采用Yolo v3 tiny模型对多尺度面目标进行检测。两种检测方法结构简单,在保持较高检测率的同时提高了运算速度,满足了工程应用中的实时性需求。4. 针对末制导阶段可能面临的自然和人为干扰,提出了一种基于多种深度特征感知的抗干扰目标跟踪方法。为了能够充分地描述目标外观,首先利用深度卷积网络中不同层输出特征图对信息描述不同的特点,提出了一种基于多种深度特征自适应选择的方法。随后,采用相关滤波的跟踪思想在连续空间域训练所需的跟踪器,提高了跟踪精度。实验结果表明,该方法能够有效应对跟踪时可能遭受的目标遮挡、形变等挑战,并展示了出色的抗干扰跟踪能力。5. 设计了空空红外成像末制导的工作流程。根据红外末制导中导引头与目标之间的距离变化规律,分析了不同阶段红外成像末制导所要解决的关键问题,并结合已经提出的红外弱小目标检测方法、多尺度目标检测方法以及抗干扰跟踪方法,设计了不同进入条件下的工作流程,对将论文研究成果向实际装备推广应用进行了初步探讨。
Other AbstractIn order to improve the ability of target detection and tracking under complex backgrounds, this paper takes air-to-air infrared imaging terminal guidance as the background, and researches on target detection and tracking technologies under complex backgrounds. The main contributions of this paper are as follows: 1. To solve the problem of insufficient detection capability of small targets under complex backgrounds, an infrared small target detection method based on the first-order directional zero-crossing detection is proposed. Based on the imaging characteristics of infrared detector at long distance, this method analyzes the imaging difference between small target and background clutters, and designs a first-order directional derivative filter, which filters the image in many different directions, and improves the signal intensity of small target. Then the filter is used to detect small targets based on the theory of zero crossing detection. Experimental results show that this method can improve the signal-to-clutter ratio of the image fully and suppress background clutter, and can detect small infrared targets under complex backgrounds accurately. 2. An infrared small target detection method based on third-order tensor RPCA is proposed. To protect the correlation between pixels, a tensor block model is proposed firstly to transfer data from two-dimensional space to three-dimensional tensor space. Then, the sparseness of the small target image and the low rank of the background clutter are used to transform the small target detection problem into a numerical optimization problem. Finally, the small target image and the background image are separated by RPCA. The experimental results show that this method can separate small target images from the original image effectively, especially in scenes with weak signal intensity and small target features which are not obvious. 3. Aiming at the problem of the changing target with multiple scales in terminal guidance, research is conducted on multi-scale point target detection and multi-scale area target detection from practical engineering applications. In terms of multi-scale point target detection, a multi-scale point target detection method based on local contrast between the central region and the neighborhood is proposed. For multi-scale area target detection, the Yolo v3 tiny model is applied to detect multi-scale area targets. The structure of these two detection methods is simple, which improves the operation speed while maintaining high detection rate, and meets the real-time requirements in engineering applications. 4. Aiming at the natural and human disturbances that the detector may suffer in terminal guidance, an anti-jamming target tracking method based on multiple depth feature perception is proposed. In order to fully describe the appearance of the target, firstly, the output feature map of different layers in the deep convolution network is used to describe different characteristics of information, and an adaptive selection method based on multiple deep features is proposed. Then, the idea of correlation filter is adopted to train the tracker in the continuous domain, which improves the tracking accuracy. The experimental results show that this method can deal with the challenges such as occlusion and deformation effectively, and show the excellent anti-jamming tracking ability. 5. The workflow of infrared terminal guidance is designed. According to the distance between the seeker and the target, the key problems need to be solved in the infrared imaging terminal guidance with different stages are analyzed. And the application scope and scenarios with each proposed method are given, including the infrared small target detection methods, multi-scale target detection methods and anti-jamming tracking methods. The workflow under different entry conditions is designed. A preliminary discussion on the application of the research results of the thesis to the actual equipment.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/27165
Collection光电信息技术研究室
Affiliation中国科学院沈阳自动化研究所
Recommended Citation
GB/T 7714
张祥越. 复杂背景下目标检测与抗干扰跟踪方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2020.
Files in This Item:
File Name/Size DocType Version Access License
复杂背景下目标检测与抗干扰跟踪方法研究.(7098KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[张祥越]'s Articles
Baidu academic
Similar articles in Baidu academic
[张祥越]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[张祥越]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.