|
|||||
|
|
| 基于模糊C均值聚类的云图样本修正与云类自动识别 | |
| 引用本文： | 王彦磊,张韧,孙照渤,牛生杰,万齐林,梁建茵.基于模糊C均值聚类的云图样本修正与云类自动识别[J].海洋科学进展,2005,23(2):219-226. |
| 作者姓名： | 王彦磊 张韧 孙照渤 牛生杰 万齐林 梁建茵 |
| 作者单位： | 1. 解放军理工大学气象学院,海洋与空间环境系,江苏,南京,211101 2. 解放军理工大学气象学院,海洋与空间环境系,江苏,南京,211101;南京信息工程大学,大气科学博士流动站,江苏,南京,210044;中国气象局,热带海洋气象研究所,广东,广州,510080 3. 南京信息工程大学,大气科学博士流动站,江苏,南京,210044 4. 中国气象局,热带海洋气象研究所,广东,广州,510080 |
| 基金项目： | 东南沿海地区云和降水监测分析基金(参气字2002第35号)，国家自然科学基金项目——西太平洋副热带高压中短期数值预报误差修正研究(40375019)，中国博士后科学基金项目(2004036012) |
| 摘 要： | 基于云类样本的红外-可见光二维灰度空间投影，采用模糊聚类方法调整优化云类样本特征区域，消除采样误差。针对常规模糊C均值聚类(FCM)方法在处理上述问题时表现出的局限性，提出用样本特征均值替代FCM中随机初始中心的改进办法，既避免了常规FCM方法对初始中心敏感的缺陷，又可纠正其聚类结果对云类样本特征结构的歪曲。改进后的聚类结果既消除了采样误差，又保持了云类样本的基本特征属性。基于该判据的分类结果，可较为准确地分辨出陆地、水体、低云、中云、卷云、对流云和积雨云，分割判另9结果符合客观实际。 |
| 关 键 词： | 模糊C均值聚类方法 云类分割识别 特征空间 |
| 文章编号： | 1671-6647(2005)02-0219-08 |
| 修稿时间： | 2004年6月3日 |
Modification of Cloud Picture Sample and Automatic Identification of Cloud Type Based on Fuzzy Clustering Method |
|
| WANG Yan-lei,ZHANG Ren,SUN Zhao-Bo,NIU Sheng-jie,Wang Qi-Lin,LIANG Jian-Yin.Modification of Cloud Picture Sample and Automatic Identification of Cloud Type Based on Fuzzy Clustering Method[J].Advances in Marine Science,2005,23(2):219-226. | |
| Authors: | WANG Yan-lei ZHANG Ren SUN Zhao-Bo NIU Sheng-jie Wang Qi-Lin LIANG Jian-Yin |
| Institution: | WANG Yan-lei 1,ZHANG Ren 1,2,3,SUN Zhao-bo 2,NIU Sheng-jie 2,WANG Qi-lin 3,LIANG Jian-yin 3 |
| Abstract: | Based on two-dimensional (infared and visible) gray space projection of cloud classification samples, the fuzzy clustering method (FCM) is used to adjust and optimize the characteristic area of cloud classification samples and to reduce the sampling errors. In view of the limitation of conventional FCM in tackling above problems, an improved FCM to use the characteristic mean of cloud samples instead of the random initial clustering center is proposed to avoid the defect to be sensitive to initial clustering center in the conventional FCM and to rectify the distortion of characteristic structure of cloud samples by the clustering results. Therefore, the improved FCM clustering results can reduce the sampling errors and retain the main attributes of cloud classification samples. The classification results can be used to correctly identify land, water, low cloud, middle cloud, cirrus, convective cloud and cumulonimbus, and the segmentation and discrimination results are consistent with the objective facts. |
| Keywords: | fuzzy clustering method (FCM) cloud segmentation and identification characteristic space |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ | |