Abstract:Cloud radiation properties and distribution significantly determine the forecasting accuracy and the climate monitoring effectiveness.Cloud detection and recognition are crucial for atmospheric sounding and atmospheric remote sensing.The purpose of this study is to realize the classification of cirrus,cumulus,stratus and clear sky by means of extracting texture features,color features and sift features to automatically train the classifier.This paper uses the extreme learning machine to study the samples and does cloud-type classification and recognition under different experimental conditions.The experiment results show that using texture features,color features and sift features together get better performance than using these features alone or any two of them together,and the accurate identification rates of cirrus,cumulus,stratus and clear sky are 87.67%,90.75%,74.50% and 93.63%,respectively,with an average of 86.64%.Under the same experiment conditions,the proposed method can outperform the artificial neutral network(ANN),the k-nearest neighbor(KNN) and the support vector machine(SVM).