I am recently working on fixing the Landsat 7 gaps by using deep learning technique. Because I found that the tools and methods provided by USGS are not very ideal for me.
Depending on your scenarios, USGS give out two solutions:
My idea is simple: use the SLC-on Landsat 7 images and Landsat 5 as training datasets to train a deep neural network (a latest one, maybe ResNet or UNet for image fusion). Use the trained network to fill the gaps in the SLC-off images.
As one of the most accurately corrected remote sensing satellite, the images of Landsat 7 have a lot to discover. But the gaps since 2003 are annoying and have been the major obstacle to let people full utilize it. Hope this method could revive Landsat 7 images in better exploring the Earth dynamics.