| 引用本文: | 欧阳彤，汪 玲，朱岱寅，李 勇. 基于LightGBM的气象目标分类技术[J]. 雷达科学与技术, 2023, 21(6): 621-629.[点击复制] |
| OU YANG Tong, WANG Ling, ZHU Daiyin, LI Yong. Meteorological Target Classification Technology Based on LightGBM[J]. Radar Science and Technology, 2023, 21(6): 621-629.[点击复制] |
|
| 摘要: |
| 为克服传统气象目标分类算法对人为设置经验参数的依赖性，本文提出一种基于轻量级梯度提升机（Light Gradient Boosting Machine，LightGBM）的气象目标分类技术。将KVNX气象雷达获取的4个极化参量（水平反射率因子、差分反射率、相关系数和差分相移率）作为气象目标的特征参数，结合参考分类标签，制作向量数据集，再进行预处理，生成满足模型需求的数据集。以此数据集为驱动，建立一种LightGBM算法的气象目标四分类模型，该模型可有效识别3种气象目标（中小雨、冰雹和湿雪）及杂波（生物杂波与地杂波）。最后，根据气象雷达观测测试数据集进行测试，结果表明该模型在有高效率识别速率条件下，识别准确率可达95%以上。再用KTLX雷达两次实际观测数据来验证模型通用性，结果表明LightGBM分类模型可有效完成4种气象目标识别，具有优越的鲁棒性。 |
| 关键词: 气象雷达 气象目标分类 轻量级梯度提升机 机器学习 |
| DOI：DOI:10.3969/j.issn.1672-2337.2023.06.005 |
| 分类号:TN959.4 |
| 基金项目:工信部民机专项资助课题（No.MJ?2018?S?28） |
|
| Meteorological Target Classification Technology Based on LightGBM |
|
OU YANG Tong, WANG Ling, ZHU Daiyin, LI Yong
|
|
Key Laboratory of Radar Imaging and Microwave Photonics, Ministry of Education, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China
|
| Abstract: |
| In order to overcome the dependence of traditional meteorological target classification algorithm on artificial setting of empirical parameters, this paper proposes a meteorological target classification technology based on light gradient boosting machine （LightGBM）. Taking four polarization parameters （horizontal reflectivity factor, differential reflectivity, correlation coefficient and differential phase shift rate） obtained by KVNX meteorological radar as the characteristic parameters of meteorological targets, and combining with reference classification labels, vector data sets are made and processed to generate data sets that meet the needs of the model. Driven by this data set, a four?classification model of meteorological targets based on LightGBM is established, which can effectively identify three kinds of meteorological targets （rain, hail and wet snow） and clutter （biological clutter and ground clutter）. Finally, the tests are made by use of the meteorological radar observation and test data set. The results show that the model has a high recognition rate and the simultaneous recognition accuracy can reach more than 95%. The universality of the model is verified by two actual observation data of KTLX radar. The results show that the LightGBM classification model can effectively identify four meteorological targets and has superior robustness. |
| Key words: weather radar meteorological target classification light gradient boosting machine （LightGBM） machine learning |
