CASIA OpenIR  > 毕业生  > 博士学位论文
基于卫星图像的台风自动检测及定位算法研究
Alternative TitleStudy on automatic typhoon detection and locating method using satellite image
刘年庆
Subtype工学博士
Thesis Advisor张文生 ; 王珏
2014-05-28
Degree Grantor中国科学院大学
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Keyword卫星图像 台风检测 台风形态分类 台风定位 特征匹配 Satellite Image Detection Of Typhoon Typhoon Morphological Classification Positioning Of Typhoon Center Feature Matching
Abstract中国位于西北太平洋的西岸,每年平均受到9~10个台风的侵袭。台风对我国东南沿海省市的人民生命和财产安全造成严重威胁,对工农业生产和交通运输造成巨大损失。正确面对台风、减少台风的灾害是一个关乎人民生命和财产安全的问题,也是防灾减灾工作的重点。准确地对台风路径进行预报是减小其造成灾害的关键,而准确预报台风路径的前提条件是能对台风进行自动检测和精确定位,这是一个国际性难题。 在各种监测台风的设备中,卫星具有覆盖范围广,资料质量高,通道数量多等优势,已成为最主要的台风监测手段。本文主要研究利用静止卫星红外图像检测台风区域,并进而确定台风中心的理论与算法,主要创新点如下: 1. 在检测台风方面,根据台风的强度和检测目标的不同,分别提出了基于梯度方向分类的台风检测算法、基于两阶段级联的台风区域检测算法。通过对2011年全年有定位数据的台风云图进行检测表明:在检测新生成的可能对我国造成影响的热带低压和热带风暴时,基于梯度方向分类的台风检测算法的误检率1.6%,漏检率15.81%;在检测强台风时,基于两阶段级联的台风区域检测算法的误检率3.10%,漏检率4.86%。通过台风检测算法,可以从整幅卫星云图中快速准确的分割出台风区域。 2. 对无历史信息的单帧卫星云图,提出了“先分类再定位”的新理念并发现了椭圆系、弯曲系两类新的台风形态(传统台风形态只有螺旋系),设计出基于趋势线编码词频的台风云系分类算法,针对椭圆系和弯曲系的台风,分别设计了不同的定位算法。 在分类方面,提出了基于趋势线编码词频的台风云系分类算法,首先提取台风云系纹理趋势线,然后对趋势线上相邻线段的角度变化进行编码,通过统计编码的分布,将台风云系自动聚类,聚类结果表明:除传统的椭圆系台风外,还有椭圆系和弯曲系两类台风形态。在定位方面,螺旋系的定位算法已经相当成熟,而椭圆系和弯曲系定位算法尚未有人研究,针对椭圆系云型,提出了基于椭圆拟合的台风定位算法,该算法通过对趋势线段进行椭圆拟合,精确确定台风中心;针对弯曲系云型,提出了基于拐点判定的台风定位算法,该算法通过寻找趋势线编码的突变来确定其拐点,将拐点位置定为台风中心。 通过利用MTSAT卫星2011年~2012年的数据进行分类算法实验,结果表明:根据本文提出的自动分类算法所分类别进行定位,超过84%的概率选择出定位误差最小的类别,说明了该分类算法有效。通过对2012年全年22个台风进行定位算法实验,结果表明:本文提出的定位算法对强热带风暴的定位误差均值为纬向0.10度,经向0.11度;对台风的定位误差均值为纬向0.05度、经向0.01度;对强台风的定位误差均值为纬向0.08度、经向0.01度;对超强台风定位误差均值为纬向0.06度、经向0.06度。 3. 对于有历史信息的台风云图,根据台风中心移动特性,提出了时空一致性特征筛选算法、均匀旋转分布特征筛选算法,实现对SIFT特征筛选,进而提出了基于时空一致性SIFT特征的台风定位算法,对当前台风与历史台风云图进行匹配,定位台风中心。 在特征筛选方面,首先,提出了时空一致性特征筛选算法,根据台风移动和旋转速度,建立了...
Other AbstractLocated in the Western North Pacific and South China Sea basin, a region where the tropical cyclone (TC) activity is frequent and strongest, China is one of the countries that are seriously affected by TCs. In fact, the country suffers from TCs about 9-10 times per year on average, which causes serious property losses, life threats, and infrastructure damages. In order to reduce the disaster of a TC, it is important to precisely forecast its path, which needs to detect the existence of TC and locate its center in advance. It is a international problem. Among all the manners of monitoring TCs, satellites have the advantages of wide cover range, high quality, and multiple channels number compared with radars and planes. Hence, satellites have become the main tool in the detection and location of TCs. In this thesis, the efficient automatic detection and location of TCs using images from infrared satellites are investigated. The main contributions of the thesis are as follows. (1) In terms of detection TC, different TC detection algorithms are proposed according to detection purposes. The algorithm based on the classification of Histogram of Oriented Gradient (HOG) is applied for detecting the formative phase of TC. The false positive rate (FPR) and false negative rate (FNR) of the algorithm are 1.6% and 15.81%, respectively. Besides, a two-stage cascading algorithm is presented for detecting strong TC. The FPR and FNR of the two-stage algorithm are 3.1% and 4.86%, respectively. As a result, the TC region can be detected rapidly and precisely using the two-stage algorithm. (2) The idea of classification is proposed for the location of TC without history path information. Different location algorithms are used for different kinds of TCs. A classification method for TCs is presented based on the Term Frequency of Trend Line Codeword (TFTLC). First, trend lines are extracted from TC images. Then, the angle differences of the neighboring segments are encoded, and TC structures are clustered according to the codeword term frequency. Different location algorithms are designed for various structures, such as ellipse and spiral line, etc. Moreover, a method for locating TC is proposed based on ellipse fitting. The trend lines are fitted using ellipse model, and the ellipse centers are then clustered. The cluster with the most ellipses area selected for the computation of TC center, which is the mean of the mean of the centers of these ellipses. For the winding ...
shelfnumXWLW1962
Other Identifier200818014628047
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/6632
Collection毕业生_博士学位论文
Recommended Citation
GB/T 7714
刘年庆. 基于卫星图像的台风自动检测及定位算法研究[D]. 中国科学院自动化研究所. 中国科学院大学,2014.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20081801462804(6606KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[刘年庆]'s Articles
Baidu academic
Similar articles in Baidu academic
[刘年庆]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[刘年庆]'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.