Geoinformatics Unit

Paper published in ISPRS International Journal of Geo-Information

September 26, 2018

Our paper entitled "Multi-temporal Sentinel-1 and -2 data fusion for optical image simulation" has been published in ISPRS International Journal of Geo-Information.

We propose a new technique based on a conditional generative adversarial network with a residual architecture to simulate optical imagery when only SAR data are available. Different from the existing techniques that use only a single SAR image as input, our approach uses a pair of SAR-optical images taken before (or after) the time of interest as input additionally. The experiments demonstrate that our approach outperforms the state-of-the-art method, producing realistic images.

More details are reported in the following open access paper:
W. He and N. Yokoya, "Multi-temporal Sentinel-1 and -2 data fusion for optical image simulation," ISPRS International Journal of Geo-Information, vol. 7, no. 10, pp. 389, 2018.

img12

The first, second, and third column images indicate SAR images acquired at time 1, optical images acquired at time 1 and time 2 (to be estimated), respectively. The fourth column images are obtained by the benchmark method, and the fifth and sixth columns are estimated by CNN and cGAN by using a single SAR image as input. The last two columns show the results obtained by our methods. The results demonstrate the benefit of using a pair of SAR and optical images taken before (or after) the time of interest as input additionally.