Research Article
Cloud Prediction Based on the Combination of Optical Flow and Deep Learning
@INPROCEEDINGS{10.1007/978-3-030-93398-2_41, author={Peng Muzi and Zhao Kanglian and Dai Zheng and Li Wenfeng}, title={Cloud Prediction Based on the Combination of Optical Flow and Deep Learning}, proceedings={Wireless and Satellite Systems. 12th EAI International Conference, WiSATS 2021, Virtual Event, China, July 31 -- August 2, 2021, Proceedings}, proceedings_a={WISATS}, year={2022}, month={1}, keywords={Laser communication Cloud prediction model Deep learning}, doi={10.1007/978-3-030-93398-2_41} }- Peng Muzi
Zhao Kanglian
Dai Zheng
Li Wenfeng
Year: 2022
Cloud Prediction Based on the Combination of Optical Flow and Deep Learning
WISATS
Springer
DOI: 10.1007/978-3-030-93398-2_41
Abstract
Satellite-to-ground laser communication has a problem of being susceptible to specific atmospheric environments, which will attenuate the laser transmission signal severely. To solve this problem, we have to know important prior information about whether the construction of a specific laser communication link is suitable. In this paper, in order to predict future images of cloud clusters around the laser links in advance, we propose a cloud prediction model based on the combination of optical flow and deep learning. Our model is based on Deep Voxel Flow (DVF), an end-to-end CNN designed for video frame synthesis. The 3D optical flow vector across space and time in the input cloud images is used to form an intermediate layer in DVF. By using DVF for multiple times to iterate the input cloud images at t second and t + 25 s, we can get the predicted cloud images during the next 100 s. Our experimental results show that, compared to the optical flow extrapolation method which is a typical method used for nowcast, our cloud prediction model can predict future cloud images with higher quality and accuracy.
