快速聚类和3阶CNN算法在雹云判别中的应用

Application of K-means clustering and CNN algorithm in hail cloud determination

冰雹是由强对流天气系统引起的,与地域存在很大的关系,地域性的判别函数构建为时下研究热点.冰雹的预测预报大多是基于雷达返回的基础数据.依据阿克苏地区气象雷达图像,采用统计学中的快速聚类算法和3阶细胞神经网络(CNN)轮廓提取算法相结合,对气象雷达图像中的云层区域进行处理,提取该云层的内外层轮廓,并计算得到内外层轮廓距离的方差.通过分析气象雷达图像的雹云图像和非雹云图像内外轮廓距离方差的区别与联系,构造判别函数.最后采用新疆阿克苏地区已有的气象雷达降雹云层图像和非降雹云层图像进行验证,验证该判别函数有效.表明结合快速聚类算法和3阶CNN轮廓提取算法得到的内外轮廓距离是一个可以有效判别是否气象雷达云层降雹的方法.

Hail were caused by strong convective weather system, great relationship with regional and regional function construction for the current research. Hail forecast most of the radar returns of the underlying data. Based on the meteorological radar images in Aksu, by using statistical fast clustering algorithm and in cellular neural networks （CNN） contour extraction algorithm, combining meteorological radar images of clouds in the area for treatment, the inner and outer contours of the clouds were extracted, and the distance between inner and outer contour variance was calculated. By analyzing the hail storm of meteorological radar images image and non-image of hail cloud profiles from the relationship between variance and structures discriminant function. Weather radar already used in Aksu Prefecture, Xinjiang hail cloud images and non-hail clouds images for authentication, it verify that the discriminant function is valid. Show that combines a fast clustering algorithm of contour extraction algorithm and CNN profiles range is an effectively to distinguish whether the meteorological radar method of hail cloud.