Feature-Enhanced Cloud Image Prediction Algorithm Based on Spatio-Temporal Attention Gated Recurrent Unit

Xiuzai Zhang; Jingxuan Li; Changjun Yang; Xuan Feng

doi:10.3788/LOP231059

Laser & Optoelectronics Progress, Volume. 60, Issue 24, 2401001(2023)

Feature-Enhanced Cloud Image Prediction Algorithm Based on Spatio-Temporal Attention Gated Recurrent Unit

Xiuzai Zhang^1,2, Jingxuan Li², Changjun Yang^3,4、*, and Xuan Feng⁵

¹Jiangsu Province Atmospheric Environment and Equipment Technology Collaborative Innovation Center, Nanjing University of Information Science & Technology, Nanjing 210044, Jiangsu, China

²School of Electronic and Information Engineering, Nanjing University of Information Science & Technology, Nanjing 210044, Jiangsu, China

³Key Laboratory of Radiometric Calibration and Validation for Environmental Satellites, National Satellite;Meteorological Center (National Center for Space Weather), China Meteorological Administration, Beijing 100081, China

⁴Innovation Center for FengYun Meteorological Satellite (FYSIC), Beijing 100081, China

⁵Technology and Engineering Center for Space Utilization, Chinese Academy of Sciences, Beijing 100094, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Basic framework of GRU

Download full size

Fig. 2. Overall framework of CrevNet

Download full size

Fig. 3. Overall framework of SmartCrevNet

Download full size

Fig. 4. Two-way SGE autoencoder

Download full size

Fig. 5. Lightweight attention model SGE

Download full size

Fig. 6. STA-GRU structure

Download full size

Fig. 7. Reversible STA-GRU module

Download full size

Fig. 8. Two prediction examples from satellite cloud images

Download full size

Fig. 9. Frame-wise MSE and SSIM comparison of the different models on the satellite cloud map dataset

Download full size

Fig. 10. Two sets of prediction examples

Download full size

Table 1. Comparative analysis of various prediction algorithms (4 frames→4 frames)

View table

Table 1. Comparative analysis of various prediction algorithms (4 frames→4 frames)

Method（4→4）	MSE / $10^{- 3}$ ↓	MAE / $10^{- 3}$ ↓	SSIM↑	PSNR↑
ConvLSTM	27.41	104.19	0.448	15.62
ConvGRU	26.81	99.07	0.444	15.72
PredRNN	25.30	97.11	0.457	15.97
PredRNN++	25.11	95.06	0.451	16.00
CrevNet	19.64	72.41	0.482	17.07
SmartCrevNet	18.21	67.34	0.520	17.40

Table 2. Two-way autoencoder and prediction module ablation experiments

View table

Table 2. Two-way autoencoder and prediction module ablation experiments

Model（4→4）	MSE / $10^{- 3}$ ）↓	MAE / $10^{- 3}$ ）↓	SSIM↑	PSNR↑
CrevNet	19.64	72.41	0.482	17.07
SmartCrevNet w/o SGE	19.47	69.36	0.518	17.11
SmartCrevNet w/o HIM	19.64	69.16	0.518	17.06
SmartCrevNet+SE+HIM	19.13	68.08	0.511	17.18
SmartCrevNet+CBAM+HIM	19.34	69.68	0.504	17.14
SmartCrevNet+SGE+HIM	18.21	67.34	0.520	17.40

Table 3. Quantitative evaluation of different methods on Moving MNIST
View table
Table 3. Quantitative evaluation of different methods on Moving MNIST
Model MSE / $10^{- 3}$ ↓ SSIM↑
ConvLSTM 11.562 0.882
PreRNN 9.496 0.905
PredRNN++ 8.344 0.917
CrevNet 5.603 0.925
SmartCrevNet 5.261 0.936

Tools

Get Citation

Copy Citation Text

Xiuzai Zhang, Jingxuan Li, Changjun Yang, Xuan Feng. Feature-Enhanced Cloud Image Prediction Algorithm Based on Spatio-Temporal Attention Gated Recurrent Unit[J]. Laser & Optoelectronics Progress, 2023, 60(24): 2401001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites