SSTSTF: A Novel Spatiotemporal Fusion Method for Remote Sensing Image Analysis

This paper introduces SSTSTF, a novel spatiotemporal fusion method for remote sensing image analysis. To verify the effectiveness of SSTSTF, we compare it with six other spatiotemporal fusion algorithms:

STARFM (Gao et al., 2006): Weighted-based* EBSCDL (Wu et al., 2015): Dictionary-learning-based* FSDAF (Zhu et al., 2016): Unmixing-based* Fit-FC (Wang and Atkinson, 2018): Weighted-based* BiaSTF (Li et al., 2020b): Deep-learning-based* GAN-STFM (Tan et al., 2021): Deep-learning-based

All competing algorithms utilize their default parameter settings.

SSTSTF's Key Feature: Separate Parameter Optimization

Unlike conventional methods, SSTSTF employs a distinct parameter setting strategy for its different components. This approach enhances the algorithm's adaptability and accuracy.

Training Process:

During the training phase, each image pair is divided into 256 x 256-sized patches with a step size of 200. This partitioning serves as input and output for the model.

Spatial Model: The Adam optimizer is used for training with B1=0.9, B2=0.99, and e=10-8. The batch size is set to 24. Training iterates for 300 epochs with an initial learning rate of 0.0001, which is decayed by 0.1 using the cosine annealing strategy.
Sensor Model: The Adam optimizer is also utilized with B1=0.9, B2=0.999, and e=10-8. The batch size is set to 36. Training runs for 300 epochs, starting with a learning rate of 0.0001. The learning rate decays by 0.1 when the loss ceases to decrease for 20 epochs.
Temporal Model: The Adam optimizer is used again with B1=0.9, B2=0.999, and e=10-8. The batch size is set to 18. Training continues for 500 epochs with an initial learning rate of 0.0002.

Conclusion

SSTSTF emerges as a robust and efficient method for spatiotemporal fusion in remote sensing image analysis. Its distinct parameter optimization strategy for different components results in superior performance compared to existing algorithms. Further research can explore the application of SSTSTF in various remote sensing applications, such as land cover mapping, disaster monitoring, and environmental assessment.

SSTSTF: A Novel Spatiotemporal Fusion Method for Remote Sensing Image Analysis