SIA OpenIR  > 光电信息技术研究室
Alternative TitleResearch on Automatic Target Recognition in Infrared Imaging Based on Geometric Primitives
Thesis Advisor罗海波
Keyword几何基元 红外成像 深度学习 匹配滤波 自动目标识别
Call NumberTP391.4/J57/2017
Degree Discipline模式识别与智能系统
Degree Name博士
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract本文面向空面成像制导应用需求,从降低信息保障要求和提高算法的普适性出发,研究基于几何基元的红外成像自动目标识别方法,该方法不依赖于导航信息支持,具有更好的普适性。本文的主要工作如下:1) 对自动目标识别技术的发展历程、应用领域以及国内外研究现状进行了总结,对本文的研究背景、技术内涵面临挑战进行了分析。制定论方案 对本文的研究背景、技术内涵面临挑战进行了分析。制定论方案 对本文的研究背景、技术内涵面临挑战进行了分析。制定论方案 对本文的研究背景、技术内涵面临挑战进行了分析。制定论方案 对本文的研究背景、技术内涵面临挑战进行了分析。制定论方案 对本文的研究背景、技术内涵面临挑战进行了分析。制定论方案 对本文的研究背景、技术内涵面临挑战进行了分析。制定论方案 对本文的研究背景、技术内涵面临挑战进行了分析。制定论方案 以及,明确了论文研究内容各章节之间的关系。2) 对几何基元进行了研究,在此础上提出一种表征方法对几何基元进行了研究,在此础上提出一种表征方法对几何基元进行了研究,在此础上提出一种表征方法对几何基元进行了研究,在此础上提出一种表征方法主要包含平面几何基元的构造、描述及提取方法 要包含平面几何基元的构造、描述及提取方法。根据。根据 仿射变换的面积不量 提 出了一种基于直线组几何元的 仿射不变量。实验 仿射不变量。实验 结果表明,本文提出的算法 能够有效地 解决目标与 待匹配图像之间存在的仿射变换 问题。3) 研究了实际光系统的成像特性,受 研究了实际光系统的成像特性,受 到通讯和雷达等电子信息处理系统中广泛应 用的匹配滤波技术启发,提出了 一种基于边缘检测方法;设计用的匹配滤波技术启发,提出了 一种基于边缘检测方法;设计用的匹配滤波技术启发,提出了 一种基于边缘检测方法;设计用于边缘提取的匹配滤波器,出了一种噪声抑制方 法和响应二值化用于边缘提取的匹配滤波器,出了一种噪声抑制方 法和响应二值化用于边缘提取的匹配滤波器,出了一种噪声抑制方 法和响应二值化用于边缘提取的匹配滤波器,出了一种噪声抑制方 法和响应二值化法;实验结果表明,本文提出的方具有更 高检测灵敏度精、法;实验结果表明,本文提出的方具有更 高检测灵敏度精、法;实验结果表明,本文提出的方具有更 高检测灵敏度精、法;实验结果表明,本文提出的方具有更 高检测灵敏度精、法;实验结果表明,本文提出的方具有更 高检测灵敏度精、强的抗噪声干扰能力以及更少时间消耗等特点。4) 研究了深 度学习原理里及其应用于边缘提取的优势和不足,出一种基研究了深 度学习原理里及其应用于边缘提取的优势和不足,出一种基度学习的边缘提取方法。对 HED HED网络结构进行了改,去掉其第三个和四 网络结构进行了改,去掉其第三个和四 个池化层,并相 个池化层,并相 应地修改了最后两个侧边输出层中的反卷积,一方面在不影 应地修改了最后两个侧边输出层中的反卷积,一方面在不影 响高层语义信息的同时提取更为精细边缘,另一方面减少了计算资源消 响高层语义信息的同时提取更为精细边缘,另一方面减少了计算资源消 响高层语义信息的同时提取更为精细边缘,另一方面减少了计算资源消 耗,有利于最终的融合。后基匹配滤波边缘提取方法结果耗,有利于最终的融合。后基匹配滤波边缘提取方法结果耗,有利于最终的融合。后基匹配滤波边缘提取方法结果耗,有利于最终的融合。后基匹配滤波边缘提取方法结果得到既能表达目标主结构,又有较高边缘聚焦力的提取果。5) 在上述研究工作的基 础,将几何元应用于异源图像匹配中提出了一种在上述研究工作的基 础,将几何元应用于异源图像匹配中提出了一种在上述研究工作的基 础,将几何元应用于异源图像匹配中提出了一种于几何基元的自动目标识别方法。首先对 直线 进行组合构成直线几何基元,然后 利用直线组中各条构成的三角形面积之比作为描述特征量。 该方法不仅可以在所有候选直 线组中找到与目标匹配的最佳线组,还能确定 线组,还能确定 组内直线的一对应关系,求出目标 组内直线的一对应关系,求出目标 模板 与实时图中的同名点。 根据 两幅图像 中同名点可以求出两幅图像之间的仿射变换矩阵,进而,进而 找出目标在红外实时图 中的位置。同时还提出了一种在实图中进行直线纯的方法。同时还提出了一种在实图中进行直线纯的方法,以提 取实时图中的主要直线,进而提高目标定位精度。
Other AbstractAutomatic target recognition is an important research direction in the field of computer vision, and it is also one of the research hotspots. Whether in the civilian or military fields, automatic target recognition technology has a wide range of applications. Automatic target recognition method based on template matching is a very important method in ATR, especially the automatic target recognition method based on heterogeneous template matching. It can be used for the guidance of air-to-surface guidance weapons, and greatly improve the air-to-surface guidance weapons. Because of its autonomy, it is of great military significance and theoretical value to carry out research on automatic target recognition based on heterogeneous template matching. In the air-to-ground imaging guidance application, the target template that can be acquired is generally a satellite image or an aerial reconnaissance image, and the real-time image is usually a forward-looking infrared image, so it is a typical problem of heterogeneous template matching. And even if an infrared template image of the target can be obtained, the infrared images acquired at different time periods and under different conditions also have great differences. Therefore, automatic target recognition based on heterogeneous template matching is a challenging task. Traditional automatic target recognition based on heterogeneous template matching requires precise navigation information to perform perspective and scale transformation on the template map. However, this method requires complex information security and is not universal enough, which limit its application in weapon models. An infrared image automatic target recognition method based on geometric primitives is proposed to meet the needs of air-to-surface imaging guidance applications, on the one hand, to reduce the information security requirements, and on the other hand, to improve the general applicability of the algorithm. This method does not depend on navigation information support and has better generality. The main work of this dissertation is as follows: 1) Summarize the development process, application fields, and research status of automatic target recognition technology, and analyze the research background, technical connotation, and challenges faced in this dissertation. The research plan for the dissertation was formulated and the relationship between the dissertation's research content and the various chapters of the dissertation was clarified. 2) The geometric primitives are studied. Based on this, a geometrical primitive is proposed for characterization, which mainly includes the construction, description and extraction methods of planar geometric primitives. According to the area invariants of affine transformation, an affine invariant based on linear geometric primitives is proposed. Experimental results show that the proposed algorithm can effectively solve the affine transformation problem between between the reference image and the image to be matched. 3) Researching the imaging characteristics of actual optical systems. Inspired by matched filtering techniques widely used in electronic information processing systems such as communications and radar, an edge detection method based on matched filtering is proposed; matching filter is designed for edge extraction, and a noise suppression method and edge response binarization method are proposed. The experimental results show that the proposed method has higher detection sensitivity, higher accuracy, stronger anti-noise interference ability and less time consumption and other characteristics. 4) We studied the advantages and disadvantages of the deep learning principle and its application to edge extraction, and proposed an edge extraction method based on deep learning. The HED network structure was improved by removing the third and fourth pooling layers and modifying the deconvolution layers in the last two side output layers. On the one hand, it extracts finer edges without affecting high-level semantic information. On the other hand, it reduces the consumption of computing resources and is conducive to the final integration. Finally, the extraction results of the edge extraction method based on matched filtering are combined to obtain the edge extraction results that can both express the target primary structure and have higher edge focusing capability. 5) Based on the above research work, geometric primitives are applied to heterogeneous image matching, and an automatic target recognition method based on geometric primitives is proposed. Firstly, the straight lines are combined to form a linear set of geometric primitives, and then the ratio of the area of the triangle formed by straight lines in the set of straight lines is used as a feature quantity describing the set of straight lines. This method not only finds the best straight line group that matches the target among all the candidate line groups, but also determines the one-to-one correspondence between the straight lines in the group and finds the corresponding points in the target template and the real-time image. According to the corresponding points in the two images, the affine transformation matrix between the two images can be obtained, and then the position of the target in the infrared real-time map can be found. At the same time, a method for straight line purification in real-time image is also proposed to extract the main straight lines in real-time iamge, thereby improving the target positioning accuracy.
Contribution Rank1
Document Type学位论文
Recommended Citation
GB/T 7714
焦安波. 基于几何基元的红外成像自动目标识别方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2018.
Files in This Item:
File Name/Size DocType Version Access License
基于几何基元的红外成像自动目标识别方法研(4779KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[焦安波]'s Articles
Baidu academic
Similar articles in Baidu academic
[焦安波]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[焦安波]'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.