首页 | 官方网站   微博 | 高级检索  
     

基于支持向量机的雷达地物回波识别研究
引用本文:魏鸣,管理,梁学伟,秦南南.基于支持向量机的雷达地物回波识别研究[J].大气科学学报,2019,42(4):631-640.
作者姓名:魏鸣  管理  梁学伟  秦南南
作者单位:南京信息工程大学 气象灾害预报预警与评估协同创新中心, 江苏 南京 210044,上海中心气象台, 上海 200030,安徽省气象台, 安徽 合肥 230031,南京信息工程大学 气象灾害预报预警与评估协同创新中心, 江苏 南京 210044
基金项目:国家重点基础研究发展计划(973计划)项目(2013CB430102);公益性行业(气象)科研专项(GYHY201306040);上海市科委资助项目(15dz1207802);国家自然科学基金青年基金资助项目(41305031);江苏省普通高校研究生科研创新计划项目(KYLX_0832)
摘    要:多普勒天气雷达探测过程中的非气象因子会显著影响雷达资料的定量化应用,在雷达基数据的应用前需对雷达资料进行抑制地物杂波、去距离折叠和退速度模糊等质量控制。本文在现有的自动识别地物回波方法的基础上,提出了基于支持向量机(Support Vector Machine,SVM)识别雷达地物杂波的方法,2013年6-8月对安庆和常州两地的CINRAD/SA雷达观测资料进行雷达地物回波识别,并将其与运用人工神经网络(Artificial Neural Networks,ANNs)识别的结果进行对比,结果表明支持向量机方法能够取得更好的效果。在地物、降水回波总样本识别和地物回波识别方面更为有效;降水回波的误判方面,神经网络略优于支持向量机,但两者差异不大,都将降水回波的误判率控制在了一个较小的范围内;另外支持向量机方法较之神经网络方法对于训练样本数目的依赖性较小,在训练样本较少时,支持向量机方法仍能保持有效的识别效果。

关 键 词:地物回波  神经网络  支持向量机
收稿时间:2017/8/15 0:00:00
修稿时间:2018/3/16 0:00:00

Ground clutter identification based on the support vector machine method with Doppler weather radar data
WEI Ming,GUAN Li,LIANG Xuewei and QIN Nannan.Ground clutter identification based on the support vector machine method with Doppler weather radar data[J].大气科学学报,2019,42(4):631-640.
Authors:WEI Ming  GUAN Li  LIANG Xuewei and QIN Nannan
Affiliation:Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters, Nanjing University of Information Science & Technology, Nanjing 210044, China,Shanghai Meteorological Center, Shanghai 200030, China,Anhui Meteorological Observation, Hefei 230031, China and Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters, Nanjing University of Information Science & Technology, Nanjing 210044, China
Abstract:Imperative quality control methods for Doppler radar data,such as ground clutter elimination,range folding elimination and velocity dealiasing,should be adopted before being used for quantitative analyses,due to the serious impacts originating from certain non-meteorological factors.In this study,in order to precisely identify the ground clutter and precipitous echo,an automatic algorithm based on the Support Vector Machine(SVM) is performed,based on the observational CINRAD/SA Doppler weather radar data in the areas of Anqing and Changzhou from June to August,2013,and the results are compared with the recognition effect based on the Artificial Neural Networks(ANNs) method.Statistical learning theory(SLT) is favorable for small samples,which focuses on the statistical law and nature of small-sample learning.As a new machine learning based on SLT,the basic principle of the SVM is to possess an optimal separating hyperplane which is able to satisfy the requirement of the classification accuracy by introducing the largest classification intervals on either side of the hyperplane.In the first step,the dataset used in the experiment will be establised by empirically distinguishing the ground clutter and precipitous points at each bin.Next,several characteristic parameters,which are used to quantify the possibility affected by the ground clutter,such as reflectivity vertical variation (GDBZ),reflectivity horizontal texture (TDBZ),velocity regional average (MDVE),and spectrum regional average (MDSW),will be derived from the reflectivity,radaial velocity,spectrum width and spatial variance information of the ground clutter and precipitous echo.The statistical results of the above characteristic parameters show the following:a large portion of these parameters vary in terms of ground clutter and precipitous echo,which indicates that the seven parameters (GDBZ,TDBZ,SPIN,SIGN,MDVE,MDSW and SDVE) contribute to the identifiable recognition of the ground clutter and precipitous echo.Based on the above conclusions,seven parameters,which are regarded as the trigger (the training factor of SVM) to establish the SVM''s training model,can be randomly extracted from the database.Finally,the training model is used to automatically recognize the ground clutter and precipitation using the random data from the database.The recognition effect of the SVM method will be examined by comparing the model output with the empirical identifications,and the examination of the ANNs algorithm is the same as that of the SVM method.The comparison of the recognition effect between the SVM and ANNs methods reveals the following:(1) The statistically identifiable recognition parameter for the sSVM and ANNs methods appears to be steady,despite the fact that the Doppler radar data vary in shape and position between Anqing and Changzhou;(2) An identifying threshold must be determined for the ANNs method before the ground clutter and precipitous echo are identified,which will lead to a differently identifiable accuracy with the unlike threshold;and (3) Overall,the SVM method works better than the ANNs method in terms of radar echo identification.Moreover,the identifiable recognition accuracy of the latter increases significantly with the increasing total number of training samples,while the identifiable recognition accuracy of the former performs at a highly accurate level,which remains relatively stable with the changes in the training samples.In terms of the identification accuracy of the total samples (ground clutter and precipitous echo) and identification accuracy of the ground clutter echo,the SVM method presents better results than the ANNs method.As for the precipitous echo erroneous recognition,the ANNs method performs slightly better than the SVM,but both methods control the erroneous recognition rate at a low level.
Keywords:ground clutter  artificial neural networks  support vector machine
点击此处可从《大气科学学报》浏览原始摘要信息
点击此处可从《大气科学学报》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司    京ICP备09084417号-23

京公网安备 11010802026262号