卫星云图云量计算是卫星气象应用的基础,现阶段对其的研究未能充分利用卫星云图的特征,导致云检测及云量计算的效果不好.针对该问题,利用多层神经网络进行卫星云图的特征提取,并通过大量实验寻找到最优的深度学习的网络结构.基于度极限学习机对卫星云图的云进行检测和分类,再利用“空间相关法”计算云图中的总云量.实验结果表明,基于传统极限学习机的深度极限学习机能够充分提取云图的特征,在进行云分类时能够较清晰地区分厚云和薄云间的界限.相比于传统阈值法、极限学习机模型以及卷积神经网络,深度极限学习机的云识别率以及云量计算准确率更高,且所提方法比卷积神经网络的效率更高.%Cloud fraction is the key point for the application of satellite imagery.The existing methods cannot make full use of characteristics of satellite imagery,resulting in ineffective cloud detection and cloud fraction.In this paper,multilayer neural network was used to extract the feature of satellite cloud image,and and through a large number of experiments,the best structure of depth learning network was found.This paper used deep extreme learning machine to detect and classify the cloud of satellite cloud image,and then used spatial correlation method to calculate the total cloud fraction.The results show that the deep extreme learning machine based on traditional extreme learning machine can extract the features of cloud images effectively,and can distinguish the boundary between thick cloud and thin cloud well.The cloud classification and cloud fraction accuracy of deep extreme learning machine are better than traditional threshold method,extreme learning machine and convolutional neural network.
展开▼
