基于卫星云图和改进AlexNet的沙尘暴预测方法

Sandstorm prediction method based on satellite cloud imageries and improved AlexNet

针对沙尘暴预测多依赖于地面气象资料,且与深度学习算法结合较少的问题,提出一种基于卫星云图和改进AlexNet的沙尘暴预测算法。首先,参照国标为地面气象资料构造沙尘暴等级标签;然后,对卫星云图进行透视变换、气象站云图数量再平衡等处理,并将卫星云图与沙尘暴等级标签相互标定;其次,改变AlexNet的卷积核尺寸,去掉局部响应归一化(LRN)层,并在所有卷积层和激活层之间增加批归一化(BN)层;最后,使用改进后的AlexNet模型进行训练、测试。实验结果表明,改进后的AlexNet模型在精确率、召回率、F1值和准确率上均优于VGGNet16、VGGNet19及AlexNet。结果表明所提方法是有效的,能为地面气象资料、卫星云图与深度学习相结合来预测沙尘暴提供参考。

Aiming at the problem that sandstorm prediction mostly relies on ground meteorological data and is less integrated with deep learning algorithms,a sandstorm prediction algorithm based on satellite cloud imagery and improved AlexNet was proposed.Firstly,classification of sandstorm intensity labels was constructed for ground meteorological data by reference to the national standards.Then,perspective transformation and meteorological station cloud imagery number rebalance were performed on the satellite cloud imageries,and the satellite cloud imageries and sandstorm intensity labels were calibrated to each other.Next,the convolution kernel size of AlexNet was changed,the Local Response Normalization(LRN)was removed,and Batch Normalization(BN)layers were added between all convolutional layers and activation layers.Finally,the improved AlexNet model was used for training and prediction.Experimental results show that the improved AlexNet model performs better than VGGNet16,VGGNet19 and the AlexNet at precision,recall,F1-score and accuracy.The results verify the effectiveness of the proposed method,which can provide references for the combination of ground meteorological data,satellite cloud imageries and machine learning methods to predict sandstorms.