Ground-based Meteorological Cloud Image Classification Based On Multiple Feature Fusion

Cloud is the external manifestation of the dynamic,thermal and water vapor cycles of the earth's atmosphere.The observation of cloud is of great significance to the study of meteorological forecast,atmospheric environment monitoring and climate change.At present,there are two main cloud observation methods,one is the satellite-based meteorological cloud observation,the other is the ground-based meteorological cloud observation.Among them,ground-based meteorological cloud observation has become one of the main means of cloud observation because of its low cost and easy operation,which can intuitively reflect the cloud information of local sky and its changing trend.Cloud shape,cloud height and cloud amount are three important parameters to be determined in ground-based meteorological cloud observation.In particular,the accurate identification of cloud shape plays an important role in the prediction of weather conditions and the evolution of the climate system.Therefore,this paper focuses on cloud recognition based on ground-based meteorological cloud images.At present,cloud recognition based on ground-based meteorological cloud images is mostly based on artificial intelligence and machine learning methods,so cloud recognition is actually an image classification task in machine learning.The so-called image classification is the use of feature extractor to extract useful information in the image(space,color,texture,etc),and then use the classifier to classify the image.Among them,feature extraction is the key.In reality,commonly used Feature extraction methods include color Feature extraction,LBP(Local Binary Pattern)Feature extraction and SIFT(Scale-invariant Feature Transform)Feature extraction,etc.However,we notice that,most of the existing feature extraction methods only extract part of the single useful feature information from the image,which leads to a low accuracy of image classification.Therefore,this paper considers the fusion of different image information extracted by different feature extraction methods for more accurate ground-based meteorological cloud image recognition.Concretely,by aggregating the color moment feature,LBP feature and SIFT feature extracted from the ground-based meteorological cloud image,this paper proposes a multi-feature fusion method for ground-based meteorological cloud image cloud recognition.It effectively integrates the global information(color moment feature)and local information(LBP feature and SIFT feature)of the cloud image to obtain a more accurate representation of the cloud image features,and improves the accuracy of cloud recognition of the ground-based meteorological cloud image.Furthermore,in order to verify the effectiveness of the ground-based meteorological cloud recognition method based on multi-feature fusion proposed in this paper,SWIMCAT(Singapore whole-sky IMaging CATegories),a ground-based weather map widely used in this fieldThe proposed method is compared with a single color moment feature,a single LBP feature,a single SIFT feature,and a method combining LBP and SIFT features.The experimental results show that the classification accuracy of the proposed method based on five-fold cross validation is better than the other methods mentioned above under three classifiers:support vector machine classifiers,decision tree classifiers and naive Bayes classifiers.