摘要
为了提高卫星云图分类精度和实时识别云类,基于云类知识库采用面向对象的分类方法对卫星云图进行分类。首先对2011年7~8月的FY-3A/VIRR卫星云图进行预处理,从中裁截500个云样本,随机选取42%云样本作为训练样本,提取训练样本的光谱和纹理特征,基于ReliefF方法进行特征选择,采用反向传播神经网络进行训练构造分类器,利用剩余58%云样本进行网络测试,至此云类知识库构建完毕。然后对待解译的云图进行JSEG分割获取云对象,基于云类知识库已训练好的分类器实现面向对象的云图分类。试验结果表明:所设计的云图分类算法有效,分类结果与云分类产品数据基本达到一致。
In order to improve the classification accuracy of satellite cloud imagery and identify cloud types in real time, a knowledge-based object-oriented classification algorithm was proposed. Firstly, 500 cloud samples were cut from FY-3A VIRR data obtained in July and August 2011 after preprocessing, and 42% of them were randomly selected for training samples. Spectral and textural features were extracted from each of training samples followed by feature selection using ReliefF, and the back-propagation neural net- work was trained to build the classifier,and tested on the rest of 58% of cloud samples,then the knowledge base of cloud types was established. To interpret a whole satellite cloud image, the JSEG segmentation method was employed to derive objects,and the trained classifier in the knowledge base of cloud types was applied to the object-oriented classification procedure. Experimental results show that the proposed algorithm for satellite cloud image classification is effective, and the generated classification results are basically consistent with the corresponding cloud classification products.
出处
《遥感技术与应用》
CSCD
北大核心
2012年第4期575-583,共9页
Remote Sensing Technology and Application
基金
国家自然科学基金资助项目(40901221)
中国博士后科学基金资助项目(20090450182)
气象灾害省部共建教育部重点实验室(南京信息工程大学)开放课题(KLME0805)
江苏高校优势学科建设工程资助项目(SZBF-2011-6-B35)