基于小样本学习的地面结露结霜现象检测方法  

Surface dew and frost phenomena detection based on small-sample learning

作　　者：朱磊[1] 张小虎 吴谨[1] ZHU Lei;ZHANG Xiaohu;WU Jin(School of Information Science and Engineering,Wuhan University of Science and Technology,Wuhan 430081,China)

机构地区：[1]武汉科技大学信息科学与工程学院

出　　处：《现代电子技术》2019年第18期130-135,共6页Modern Electronics Technique

基　　金：国家自然科学基金项目(61502358)~~

摘　　要：在《地面气象观测规范》中,地面结露、结霜现象的观测是一项重要的项目.针对当前该项目仍处于需要人工观测的情况,利用结露和结霜发生时的成像信息,提出一种基于特征学习的结露和结霜两类现象的检测和分类方法.首先,通过提取多尺度的结构、纹理和颜色特征以形成对图像的语义描述,再对提取的特征采用Fisher向量编码器以扩充特征空间,并学习一个线性的支撑向量机作为最终的分类模型.与当前热门的、基于深度卷积网络的方法相比,该方法能够在极小样本规模的条件下取得高于深度模型的分类正确率.在国内多个地面观测站点获取的地面结露和结霜图像所制备的数据集上的测试结果表明,所提方法的正确率达到了80%以上.Observation of ground surface dew and frost phenomena is an significant subject in "Specifications for Surface Meteorological Observation". In allusion to the status quo that both phenomena still rely on manual observation,a detection and classification method based on feature learning for dew-forming and frost-forming phenomena is proposed,in which imaging information during occurrence of dew and frost is adopted. The multi-scale structure,texture and color features are extracted to form semantic representation of images,and then the Fisher vector encoder is used for the extracted features to expand feature space, and a linear SVM(support vector machine)is learnt as the final classification model. In comparison with the popular methods based on deep convolutional networks,the proposed method can be used to obtain higher classification accuracy than that of deep model under the condition of minimal sample size. Additionally,the test results on the data sets of surface dew and frost images obtained from several surface observation sites in china show that the classification accuracy of the proposed method is up to 80%.

关 键 词：地面气象观测 结露现象检测 结霜现象检测 特征提取 语义描述 图像分类 

分 类 号：TN911.23.34[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]