论文部分内容阅读
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 images and identify clouds in real time, cloud-based knowledge base uses object-oriented classification to classify satellite cloud images. First of all, we preprocessed the FY-3A / VIRR satellite image from July to August 2011, cut 500 cloud samples, randomly selected 42% of cloud samples as training samples, and extracted the spectral and texture features of training samples. Based on the ReliefF method The feature selection is carried out. The back propagation neural network is used to train the classifier, and the remaining 58% cloud samples are used for network testing. At this point, the cloud knowledge base is constructed. Then, the cloud image to be interpreted is subjected to JSEG segmentation to obtain the cloud object, and an object-oriented cloud image classification is performed based on the classifier trained by the cloud knowledge base. The experimental results show that the designed cloud graph classification algorithm is effective, and the classification result and cloud classification product data basically reach the same.