【作者】 满国晶；

【导师】 钱华明；

【作者基本信息】 哈尔滨工程大学 ， 导航、制导与控制， 2011， 硕士

【摘要】 台风(热带气旋)是危害人类最严重的自然灾害之一,它所引起的狂风、暴雨及风暴潮等给沿海人们的生产生活带来了极大的不便和危害,同时也严重威胁到航海的安全。为了有效地降低台风对人们造成的影响,加强对台风的实时监测定位和预报就显得尤为重要。由于台风中心定位的好坏取决于台风云系的分割效果的好坏,因此对台风云系进行分割具有很高的研究价值和极其重要的现实意义。蚁群算法作为一种新型的模拟进化算法,其良好的离散性、并行性、正反馈性和鲁棒性,使其在图像分割领域得到了广泛的应用。本文选用风云二号气象卫星拍摄的云图作为客观的资料,以2005年9月1日14时30分在福建莆田平海镇登陆的第13号台风——泰利为例,利用蚁群算法及其改进对台风云系进行分割研究,本文主要围绕以下四个方面进行研究。论文首先对蚁群算法和图像分割方法近年来的研究进展进行了分析与总结。详细介绍了蚁群算法的基本原理及数学模型,并以TSP问题为例对蚁群算法的实现过程进行了说明。分析蚁群算法的优缺点,给出了蚁群算法在图像分割领域的一些改进措施。其次在蚁群算法的基础上,简单介绍聚类分析的相关概念及数学描述。总结近年来有代表性的四种蚁群聚类算法,并对这几种蚁群聚类算法的基本模型及其优缺点进行介绍和分析,以基于觅食原理的蚁群聚类算法为研究重点,进行台风区域的分割研究。根据数字图像的离散性特点,从聚类角度出发,以灰度为特征,将蚁群聚类算法引入到台风云系分割中,并通过引入初始聚类中心和引导函数来解决传统蚁群聚类算法计算量大,搜索时间过长的问题。在蚁群聚类算法的基础上,针对单纯采用蚁群聚类算法可能会将类似于台风云系的分布不均匀的无关云团分割出来而造成误检的情况,提出了一种蚁群聚类算法融合数学形态学方法的台风云系分割方法。鉴于台风图像是自然纹理图像,论文在灰度特征的基础上,利用图像的高层次特征——纹理特征来全面地描述图像信息,以求获得更好的分割效果。通过对现有典型纹理分割方法的分析和比较,选用灰度共生矩阵法提取台风纹理特征,形成特征向量,并以其为特征,利用蚁群聚类算法进行纹理分割,给出了一种基于灰度共生矩阵的台风图像分割方法。最后采用MATLAB7.1作为编程工具,对本文算法的有效性和准确性进行仿真验证。实验结果表明,本文的算法能够有效地对台风云系进行分割及特征提取,进而能很好的实现对台风中心定位,其研究成果对台风云系的识别、预报及定位等具有一定的现实意义。更多还原

【Abstract】 Typhoon (tropical cyclone) is one of the worst natural disasters that seriously influence human being. Strong winds, heavy rain, storm surge and other disasters caused by Typhoon have brought great inconvenience and danger to people’s production and life nearby southeast of sea in our country, even seriously threaten to the sailing security. In order to reduce disasters effectively caused by typhoon, reinforcing the real-time monitoring, location and forecasting for the typhoon is particularly important. Due to accuracy of the typhoon centre locating mainly depends on the segmentation effect of Typhoon. Therefore, the typhoon segmentation was highly valuable to study and extremely important practical significance.Ant colony algorithm is a new simulated evolutionary algorithm. Its characters of discreteness, parallelism, positive feedback and robustness make it get widely application in the image segmentation field. This paper selects FY-2 meteorological satellite cloud pictures as subject investigated,13th typhoon—TALIM as an example which landed PingHai town, PuTian city, FuJian province at 14:30 of September 1st 2005. Ant colony algorithm and its improvement are applied to the research of the typhoon segmentation, and this paper mainly involves the following four aspects of works:Firstly the research progress of ant colony algorithm and image segmentation in recent years is summarized. The principle and mathematical model of ant colony algorithm is deeply analyzed and elucidated. The realization process of ant colony algorithm is explained by solving the TSP problem. Finally, some improved algorithms of ant colony algorithm widely used in the image segmentation field are quoted.Secondly the related concepts and mathematical description of cluster analysis is briefly illuminated, and then some representative ant colony clustering algorithms in recent years are summarized. After that, four basic models are introduced briefly and merits and drawbacks of ACCA are compared. Ant colony clustering algorithm which based on the principle of foraging is used to segment the typhoon image.According to the discrete characteristics of digital images, the ant colony clustering algorithm is applied to the segmentation of the typhoon from view of the clustering. By introduced the initial clustering centers and guidance function, the large calculation and the search time of traditional ant colony clustering algorithm can be shorten obviously. The problem of low accuracy may be caused by the uneven distribution of unrelated clouds which is similar to typhoon when the ant colony algorithm is utilized respectively. To solve the problem, an ant colony algorithm integrated with mathematical morphology of typhoon segmentation is presented.Considering the typhoon image is a natural texture image, based on the grayscale characteristics, using the high-level image features--texture features can describe fully the image information for better segmentation results. Texture features of typhoon image are extracted by using GLCM and the feature vector formed.Then the typhoon-cloudy is separated from motley cloudy by using ant colony clustering algorithm.By using MATLAB7.1 as a programming tool, the effectively and accuracy of algorithm has to be tested and verified. Experiments also show that the algorithm of this article is effective to segment the typhoon-cloudy and extract the feature of typhoon area, and finally realize the location of typhoon center. The result of its research has certain practical significance for identification, forecasting and location of typhoon.更多还原