Generate DSM/DEM from multiple PlanetScope images

  • 23 November 2022
  • 6 replies
  • 132 views

Badge +1

Hello,

I am study the topography change of mining site in Vietnam and wonder is there anyone experienced the DSM/DEM generation from multiple PlanetScope images? And comparison between stereo and multiple-images result for DEM/DSM. Although there are several scientific papers, there is lack of guideline to retrieve the high accuracy of DEM/DSM. 

 

Thank you.


6 replies

Userlevel 3
Badge +3

Hey @Minh Nguyen - I’m really interested in this topic, and am starting to try to ask around to see if there’s any open source code that can be used. I was just referred to one of the papers doing this (https://www.tandfonline.com/doi/full/10.1080/15481603.2022.2060595), and will try to get in touch with the authors to see if they might share more of how they did it. Do you have links to the other papers that have done it? I’m happy to try to network to them to see if they’d be willing to share source code and/or guidelines of how they did it. 

Userlevel 3
Badge +3

So the author of that paper has some of his code up at https://u.osu.edu/qin.324/rsp/. Looks like you can apply for a test license and make use of it for academic purposes. 

Badge +1

Hi @cholmes,

Thanks a lot for this information. It sounds the user community of this software/code is not very active in the support group.

Meanwhile I’m able to generate the DSM/DEM using other software and the result looks promising in comparison with open DEMs. I guess the rest of work is verifying results and optimizing the processing flow for PlanetScope images. I may try soon with RapidEye since I reached downloading quota in the Planet Explorer.

The difference in generating DSM/DEM from PlanetScope and RapidEye is lower spatial resolution and the orbit/geometry of the images not facilitating the general algorithms. That’s why we do not see many tutorial on the community.

Anyway, I will update the potential workflow soon.

Userlevel 3
Badge +3

That’s great @Minh Nguyen. Can you share what software you used to generate your results? Would be great to share any code and workflows to help others reach similar results. 

I’m also curious about what type of data you’d need to do further experiments? We’re working on a fully open STAC data catalog, available at https://planet.com/data/stac (see STACIndex for a nicer UI to the data). I’m hoping to put more planetscope data in, and would be open to putting in some that lets people experiment with DSM/DEM creation.

There’s also some SkySat imagery in there, with stereo in some interesting high-relief areas. We could pretty easily put out some more raw asset types. I’m close to getting up some Basic L1A All Frames, assets, which I think could be used. 

Badge +1

That’s great @Minh Nguyen. Can you share what software you used to generate your results? Would be great to share any code and workflows to help others reach similar results. 

I’m also curious about what type of data you’d need to do further experiments? We’re working on a fully open STAC data catalog, available at https://planet.com/data/stac (see STACIndex for a nicer UI to the data). I’m hoping to put more planetscope data in, and would be open to putting in some that lets people experiment with DSM/DEM creation.

There’s also some SkySat imagery in there, with stereo in some interesting high-relief areas. We could pretty easily put out some more raw asset types. I’m close to getting up some Basic L1A All Frames, assets, which I think could be used. 

That’s very nice information to me, @cholmes.

I’m using Agisoft Metashape and below is basic workflow that I’m using to generate DSM/DEM from a pair of Planetscope. The advantage is that Agisoft provides GUI and friendly environment to newbie like me. Processing steps:

  1. Import raw Planetscope images.
  2. Images alignment
  3. Build dense cloud
  4. Build DEM

These steps do not guarantee to work for every pair so I choose the best one. Since the Planetscope does not include panchromatic-band so I try between bands (channels) and the results are slight different. I’m thinking to combine/merge results between channels to improve the final DSM/DEM.

The initial idea is to use multi-image (like UAV images) to generate DSM/DEM. However, I haven’t provide good enough results and current looking around for improvement. Furthermore, I prefer to go back in time to utilize RapidEye as we know the benefit of very high spatio-temporal coverage of planetscope. However, the raw image is provided with full frame so the downloading quota is reached very soon.

I’m wondering that the ASM is now able to process Planetscope/RapidEye and completely provide DSM/DEM?

Hope to learn more from your team.

Userlevel 3
Badge +3

Thanks for sharing @Minh Nguyen! Agisoft Metashape does seem like a good solution, I’ve heard great things about their accuracy, at least with drone imagery.  Definitely let us know if you see success with it, unfortunately I don’t know of anyone at Planet who has worked with it.

@joe.mascaro just pointed me at another paper and software that might be interesting as well. See https://ieeexplore.ieee.org/document/9925153 for the paper, and the software is http://www.tectonics.caltech.edu/slip_history/spot_coseis/download_software.html

One other piece of software worth mentioning is the Ames Stereo Pipeline (repo on github). There’s been good success using it for SkySat data, see this paper and accompanying repo. But I don’t (yet) know of anyone who has used it for Planetscope data. 

Reply