SIA OpenIR  > 光电信息技术研究室
近海面目标自动检测与识别方法研究
Alternative TitleResearch on Automatic Detection and Recognition of Near Sea Objects
杨雨涵
Department光电信息技术研究室
Thesis Advisor惠斌
Keyword水面无人艇 目标检测 海天线提取 特征提取
Pages83页
Degree Discipline模式识别与智能系统
Degree Name硕士
2020-05-26
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract近年来,随着人工智能的快速发展,能够实现海面目标自动检测与识别、智能路径规划和自主避障等功能的水面无人艇(Unmanned Surface Vessel,USV)在民用与军事等领域的应用越来越广泛,基于光电体制的海面目标自动检测与识别技术是实现无人艇智能化的关键技术,受到了研究人员的广泛关注。不同于陆面目标的检测和识别,在海洋环境中,无规则变化的浪花、大量的海杂波、岛岸建筑以及光照的干扰等因素都给海面目标的检测与识别带来许多挑战,目前,海面目标的自动检测与识别仍然存在许多问题。本文以水面无人艇上的光电载荷为研究背景,提出了一个新的近海面目标自动检测与识别系统,按照检测与识别系统的流程划分,研究内容包括以下几方面:1)本文调研分析了现有的海天线检测算法,并在实际海面环境中,实验对比了现有算法的性能,最终选取基于Hough变换的检测算法作为确定海天线的主要技术。为了产生更多更加可靠的海天线候选直线,本文提出在图像预处理阶段,首先采用多尺度中值滤波平滑图像,后续结合提取的候选直线的长度和颜色矩特征对候选直线进行筛选,以排除海杂波、不均匀光照和海岸等众多干扰因素。通过实验证明,本文加以改进的基于Hough变换的提取算法能更加准确地定位海天线。(2)海面目标检测环节,本文提出的基于高斯混合模型(Gaussian Mixture Model,简称GMM)的检测算法,根据序列图像的变化,构建高斯混合模型模拟多变的海上场景,通过背景减除法确定目标区域。算法首先基于上一节提取的海天线对输入图像进行校正,然后利用三帧差分算法实现背景显露区分离,随后采用不同的更新机制完成高斯混合模型的实时背景更新。最后针对舰船目标,为了进一步提高其检测的准确率,采用形态学滤波并且结合运动持久性滤波对舰船尾迹进行滤除。通过实验证明,和目前其他的海面目标检测算法对比,本文提出的算法确定舰船目标所在区域的准确程度更高,鲁棒性能更强。(3)海面目标分类识别环节,本文研究分析了海面目标的特点,选取了几何特征和Hu不变矩特征作为描述海面目标的主要特征,并且通过实验验证了用各个特征来区分海面目标的可行性。本文采用SVM分类器对上一步检测到的目标进行分类识别,实验表明,SVM分类器可以准确地识别不同种类的海面目标,取得了较为满意的分类结果。
Other AbstractWith the rapid development of artificial intelligence for the past few years, unmanned surface vessel(USV) can realize the functions of detecting and recognizing automatically maritime objects, planning path intelligently and avoiding obstacle autonomously, it is more and more widely used in civil and military fields. The automatic detection and recognition technology of maritime objects based on photoelectric system is the key technology to realize the intelligent unmanned surface vessel, which has been widely concerned by researchers. Different from the detection and recognition of land objects, in the maritime environment, many interference factors bring many challenges to the detection and recognition of maritime objects, such as irregular waves, a large number of sea clutter, shore buildings and illumination. At present, there are still many problems in the automatic detection and recognition of maritime objects. In this paper, a new automatic detection and recognition system of maritime objects is proposed based on the research background of photoelectric load on unmanned surface vessel. According to the flow of detection and recognition system, the research content includes the following aspects: (1) In this paper, the existing sea horizon line detection algorithm is investigated and analyzed. In the actual maritime environment, the performance of the existing algorithm is compared by experiment. Finally, the line detection algorithm based on Hough transform is selected as the main technology to extract horizon line. In order to produce more reliable horizon candidate lines, this paper proposes that in the image pre-processing stage, multi-scale median filter is used to smooth the image firstly, and then combines with the length feature and color moment feature of the extracted lines to screen the candidate lines, so as to eliminate many interference factors such as sea clutter, uneven illumination and island. Experiments show that the improved algorithm based on Hough transform can locate the horizon more accurately. (2) In the process of maritime objects detection, this paper proposes a detection algorithm based on Gaussian mixture model (GMM). According to the change of sequence image, the algorithm constructs Gaussian mixture model to simulate the changeable maritime scene, and determines the objects area by background subtraction. The algorithm registers the input image using the horizon line extracted in the previous section firstly, then uses three frames difference algorithm to separate the background exposure area, and uses different updating mechanisms to update the background of Gaussian mixture model in real time. Finally, in order to further improve the accuracy of ship objects detection, the ship wake is filtered by morphological filtering and motion persistence filtering. Experiments show that compared with other maritime objects detection algorithms, the algorithm proposed in this paper is more accurate and robust in determining the area where the ship is located. (3) In the process of maritime objects classification and recognition, this paper studies and analyzes the characteristics of sea objects, selects the geometric features and Hu invariant moment features as the main features to describe sea objects, and verifies the feasibility of each feature to distinguish maritime objects through experiments. In this paper, SVM classifier is used to classify and recognize the objects detected in the previous step. The experiment shows that SVM classifier can accurately recognize different kinds of maritime objects and achieve satisfactory classification results.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/27123
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
杨雨涵. 近海面目标自动检测与识别方法研究[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
近海面目标自动检测与识别方法研究.pdf(1847KB)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
[杨雨涵]'s Articles
Baidu academic
Similar articles in Baidu academic
[杨雨涵]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[杨雨涵]'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.