Cloud Fraction Of Satellite Based On Active Online Extreme Learning Machine

Cloud Fraction is not only an important parameter affecting the radiation balance of the gas system, but also an important indicator of atmospheric circulation and climate change. Cloud fraction computing is closely related to cloud detection.The accuracy of satellite cloud image classification method directly affects the accuracy of cloud computing. In the practical application of the satellite cloud image detection and processing, the expansion of the training set is one of the ways to improve the classification accuracy. However,a large number of labeled data sets need to spend a lot of manpower and material costs. In the field of remote sensing,the rapid development of modern high-resolution sensor technology makes it easy to collect unlabeled data and economy. Therefore,it is significant to improve the detection performance of the algorithm by a small number of labeled training samples and a large number of unlabeled samples.Based on the theory of machine learning, this paper combines the active learning with the extreme learning machine to fully excavate the useful information of a large number of samples in the satellite cloud image classification,and use a small number of labeled samples to improve the performance of the classifier,improve the precision of the test and reduce the artificial mark cost.The main work of this paper is as follows: (1) To study the sample uncertainty sampling evaluation strategy of the extreme learning machine, which is used in the active online extreme learning machine . Through the extreme learning machine, the active support vector machine and the active extreme learning machine in four different the performance of the public data shows the validity of the proposed active online extreme learning machine. (2) Using the active online extreme learning machine for cloud detection. The original satellite cloud samples were sampled and pretreated. Extreme learning machine as a basic classifier, using uncertainty sampling to extract information-rich samples. Without reducing the performance of the classifier under the premise of reducing the cost of manual labeling, reduce the classifier training time. By comparing with the threshold method, active support vector machine and ELM experiment, we can verify the effectiveness of the proposed method in dealing with satellite cloud data. (3) After the detection of the satellite cloud map, the use of "spatial correlation" cloud detection based on cloud computing and compared with four different algorithms. Finally, we compare and analyze the standard database with expert calibration to improve and perfect the cloud fraction computing model.