基于多特征融合的低空风切变类型识别  

Recognition of low-level wind shear type based on multi-feature fusion

作　　者：蒋立辉[1,2] 杨浩广[1] 庄子波[2] 熊兴隆[1] 

机构地区：[1]中国民航大学天津市智能信号与图像处理重点实验室,天津300300 [2]中国民航大学民航气象研究所,天津300300

出　　处：《计算机工程与设计》2015年第6期1555-1559,1564,共6页Computer Engineering and Design

基　　金：国家973重点基础研究发展计划基金项目(2010CB731800);国家自然科学基金项目(41075013);国家自然科学基金民航联合重点基金项目(U1433202);中央高校基金项目(ZXH2010D020;3122013P009)

摘　　要：针对微下击暴流、低空急流、顺逆风和侧风4种不同低空风切变的激光雷达扫描图像,提出一种基于形状特征和纹理特征相结合的识别方法。采用Zernike矩和旋转不变统一模式的局部二值模式(LBP),分别提取反映风场全局变化的形状特征和反映风场局部变化的纹理特征;将两种特征串联融合后,通过主成份分析(PCA)对其进行降维,提取有效特征;利用k近邻分类器对4种低空风切变图像进行分类。实验结果表明,与其它多种算法相比,该算法平均识别率最高,识别效果更加稳定。For four different low‐level wind shears of laser radar scan images which are microburst ,low‐level jet stream ,head and tail wind shear and side wind shear ,a recognition method based on shape and texture features was proposed .Firstly ,shape features reflecting the global changes of wind field and texture features reflecting the local changes of wind field were extracted using Zernike moments and rotation invariant uniform local binary pattern （LBP） .Then ,the two features were combined in se‐ries and the combined features were reduced by principal component analysis （PCA） to get the effective features .Finally ,k‐nea‐rest neighbor classifier was used to classify four types of low‐level wind shear images .The experimental results show that com‐pared with other algorithms ,the proposed algorithm has the highest average recognition rate and it is more stable .So this algo‐rithm can identify four low‐level wind shears effectively .

关 键 词：低空风切变 ZERNIKE矩 局部二值模式 (LBP) 主成份分析 (PCA) K近邻分类器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]