摘要
为解决对流性天气系统影响对流云区域的自动识别及分割问题,将深度学习理论用于天气雷达探测特征天气系统领域,提出一种对流云区识别算法。首先,将雷达基数据坐标转换算法以极坐标存储的原始数据转化为地理坐标,利用四线插值算法补足探测盲点,建立完整的三维数据场,结合算法需要生成垂直最大回波强度显示分析产品,标注后形成实验数据集;其次,建立U-Net网络架构的算法模型,经过训练和调参的实验过程,找到识别效果最佳的模型网络组成;最后,通过系列评价指标获取识别分割效果的统计数值支撑。实验结果表明,文中方法能够准确地识别并分割出对流云区域,特别是相对面积较大的区域。
To solve the problem of automatic recognition and segmentation of convective cloud regions affected by convective weather systems,deep learning theory is applied to the field of weather radar detection feature weather systems,and a convective cloud region recognition algorithm is proposed.The raw data stored in polar coordinates is converted into geographic coordinates by means of the radar based data coordinate conversion algorithm,and the four line interpolation algorithm is used to fill the detection blind spots,and establish a complete 3D data field.Combined with the algorithm requirements,a vertical maximum echo intensity is generated to display and analysis product,label it,and form a practical dataset.An algorithm model is established for U-Net network architecture,and the model network composition with the best recognition effect is found by means of the experimental process of training and parameter adjustment.The statistical numerical support for identifying segmentation effects is obtained by means of a series of evaluation indicators.The experimental results show that the proposed method can accurately identify and segment convective cloud regions,especially those with relatively large areas.
作者
闫军
王新舒
韩旭日
YAN Jun;WANG Xinshu;HAN Xuri(Inner Mongolia Institute of Meteorological Science,Hohhot 010051,China)
出处
《现代电子技术》
2023年第24期147-152,共6页
Modern Electronics Technique
基金
内蒙古自然科学基金项目(2019MS04002)
内蒙古科技创新引导项目(KCBJ2018006)
内蒙古自治区气象局科技创新项目(nmqxkjcx202331)。
