摘要
基于卷积计算技术,对常规卷积方法进行了改进,实现利用自定义卷积核快速识别雷达回波图像特征,并对非气象回波进行剔除操作。通过构造孤立点状、线状等卷积核,对南昌市雷达基本反射率因子回波图像和江西省雷达组合反射率因子回波拼图进行业务应用后表明:基于改进卷积方法和自定义孤立点状卷积核、线状等宽卷积核、射线状卷积核和弱回波卷积核,可以实现对雷达单站回波、组合反射率拼图中的孤立点回波、线状回波、地物回波等非气象类回波和弱回波区域的快速剔除,质量控制效果较好。其中孤立点卷积核能剔除长宽均不超过3个像素的点状回波,等宽线状卷积核能剔除水平、垂直和倾斜等横截面不超过3个像素的线性回波,射线状卷积核能剔除横截面不超过11个像素的射线状回波,弱回波卷积核能剔除成片的地物回波和弱回波等。
Based on the principle of convolution calculation, this study improves the conventional convolution method and constructs the isolated point convolution kernel, linear convolution kernel, and weak echo convolution kernel. Based on this improved conventional convolution method and these new kernels, the non-meteorological echoes and the areas with weak echo, such as isolated point echo, linear echo and ground object echo in single radar station, and combined reflectivity puzzle echo images, can be quickly eliminated. The isolated point convolution kernel can eliminate the point echo with the length and width no more than 3 pixels. The equal width linear convolution kernel can eliminate the linear echo with the horizontal, vertical and inclined cross section no more than 3 pixels. The radial convolution can eliminate the radial echo with the cross section no more than 11 pixels. The weak echo convolution kernel can eliminate the ground object echo and weak echo.
作者
聂道洋
肖安
夏侯杰
NIE Daoyang;XIAO An;XIA Houjie(Fuzhou Meteorological Bureau of Jiangxi Province,Fuzhou 344000;Jiangxi Meteorological Observatory,Weather Forecast open laboratory of Jiangxi Meteorological Bureau,Nanchang 330096)
出处
《暴雨灾害》
2022年第5期598-606,共9页
Torrential Rain and Disasters
基金
中国气象局创新发展专项(CXFZ2021Z012)
江西省气象局重点项目(JX2020Z04)
江西省气象局面上项目(JX2020M19)
抚州市气象局自筹经费科研项目(FZQXKY202006)。
关键词
卷积
卷积核
雷达回波
质量控制
convolution
convolution kernel
radar echo
quality control
