Cloudage Detection And Cloud-type Recognition Based On Ground-based Cloud Image

Cloud observation is an important content of the meteorological work, and correct identification of the basic elements of clouds such as cloud cover, type, height has important guiding significance to current weather systems. The formation of cloud, development and evolution are also the important theory basis of weather forecast in the future. There are mainly two kinds of observation methods:the observation of satellite cloud imagery and ground-based cloud image. Ground-based observation can not only describe a small range of clouds, provide local information, but also has the advantages of easy to operate, low cost, and abundant image information. Therefore, the observation technology of ground-based cloud images has become an hot research topic in the field of meteorology.This article aims to the research of ground-based cloud classification and identification, and we proposed two novel methods:one is cloud classification based on global features, local features and extreme learning machine(ELM), the other is cloud recognition based on gaussian mixture model(GMM) and support vector machine(SVM). The main results are as follows:(1) Research of cloud features which can represent cloudform, then we proposed the descriptive manner combined with global feature and local feature. Global features include: GLCM and Tamura texture feature, color histogram and color moment. Bag of words model is used to deal with SIFT feature, and the result is considered as the local feature. Finally, we adopt ELM model to study the cloud classifier. The experimental results indicate that our method can effectively improve the performance of cloud classification compared to traditional methods.(2)We propose a new method for automatic detection and recognition based on GMM and SVM. Firstly, GMM is used to simulate the probability distribution of every pixel whose parameters can be optimized by EM algorithm. After that, all the pixels can be detected by Bayesian classifier. We can extract the SIFT descriptors from the detected cloud areas and then the descriptors were processed by the Bag of Words model to generate the Bag of Word(BOW) descriptors. Finally, we use SVM to realize the recognition of cloud image. The experimental results shows that the detection precision and classification accuracy of this method have been significantly improved as compared to the traditional methods.