云导风计算中结合局部特征的区域匹配算法研究

发布时间：2018-05-04 10:08

本文选题：云导风 + 局部特征提取　；参考：《山东大学》2017年硕士论文

【摘要】：气象数值预报,是一个与科技民生息息相关的领域,随着气象卫星技术的不断进步,得到的资料也越来越丰富,气象数值预报也从依赖专业人员的经验,转而利用各类图像处理技术,使结果更加客观,更加精准。云导风技术主要指利用连续时序的卫星云图,捕捉云图上示踪云团的运动轨迹以反演风场的过程,是气象数值预报的一个重要信息来源,尤其是对于较为偏远且气象监控站点稀缺的地区。目前主流方法是基于相关度计算的模板匹配方法。然而云团的运动属于非刚性物体的半流体运动,目前来讲,并没有一个相对成熟的计算模型,现有的计算方法主要利用区域匹配,对模板在搜索区域内进行相关度计算得到云团位移,并没有充分考虑云团特征,对于有旋风带来的云团旋转并不能够很好的适应,并存在搜索方式计算量较大、对人工依然有很大依赖等缺点,因此该方向依然有很大的挖掘空间和探索价值。本文为了更好地捕捉云团运动规律,提高云导风计算效率,考虑将SIFT算法得到的特征点在前后云图上的分布规律作为云团局部特征信息,并结合到模板匹配中,是对云导风技术的一个新的尝试。SIFT算法可以有效克服尺度变化、旋转、亮度变化等影响,因此,由SIFT算法产生的特征点可以代表云团运动并对云团局部特征进行描述。在本文中,为了避免由于云团不确定运动带来的干扰,本算法没有直接选用特征点的匹配距离作为移动矢量,而是以特征点分布情况作为云团的局部特征代表,将模板匹配与局部特征结合,一方面可以避免对于云团局部存在平滑区域所造成的特征点难以检测,另一方面利用局部特征区域直接匹配,有效缩减模板匹配时遍历搜索过程的计算量。实验结果表明,基于局部特征的区域匹配算法能够适应云团的各类变化,也能够基于特征点分布得到一定的有旋风旋转信息,并且相较于传统算法能够获取更多的风矢量信息。尤其对于高分辨率的卫星图像,在保证一定准确性的同时,能够使计算效率大幅提升。
[Abstract]:The meteorological numerical forecast is a field closely related to the people's livelihood of science and technology. With the continuous progress of meteorological satellite technology, the data obtained are more and more abundant. The weather numerical forecast also depends on the experience of professionals. Instead, various image processing techniques are used to make the results more objective and accurate. Cloud wind guide technology mainly refers to the process of retrieving the wind field by capturing the track track of the cloud cluster on the continuous time series satellite cloud image, which is an important source of information for the meteorological numerical forecast. Especially for the more remote and scarce meteorological monitoring sites. At present, the main method is template matching based on correlation calculation. However, the motion of cloud cluster belongs to the semi-fluid motion of non-rigid object. At present, there is not a relatively mature calculation model. When calculating the correlation degree of the template in the search area, we can get the cloud cluster displacement without fully considering the cloud cluster characteristics, which can not be well adapted to the whirlwind cloud rotation, and there is a large amount of calculation in the search mode. There is still a great dependence on labor, so this direction still has great space and exploration value. In this paper, in order to better capture the cloud motion law and improve the efficiency of cloud wind guide calculation, we consider the distribution of the feature points on the front and rear cloud images obtained by SIFT algorithm as the local feature information of the cloud cluster, and combine it with template matching. SIFT algorithm can effectively overcome the influence of scale change, rotation, brightness change and so on. Therefore, the feature points generated by SIFT algorithm can represent the cloud movement and describe the local features of the cloud cluster. In this paper, in order to avoid the interference caused by the uncertain motion of the cloud, the matching distance of the feature points is not directly selected as the moving vector, but the distribution of the feature points is taken as the local feature of the cloud cluster. Combining template matching with local feature, on the one hand, it can avoid the difficulty of detecting feature points caused by the existence of smooth region in cloud cluster, on the other hand, it can use local feature region to match directly. Effectively reduces the computational cost of traversing the search process when template matching is performed. The experimental results show that the region matching algorithm based on local features can adapt to all kinds of changes of cloud clusters, and can obtain some whirlwind rotation information based on the distribution of feature points, and can obtain more wind vector information than the traditional algorithm. Especially for high resolution satellite images, the computational efficiency can be greatly improved while ensuring certain accuracy.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P456.7;TP391.41

【参考文献】