5 ms·

Are you saying there are big innovations in the optics for the newer ones? I mean, we saw what the current capability was last year, and it that didn't seem li

Are you saying there are big innovations in the optics for the newer ones? I mean, we saw what the current capability was last year, and it that didn't seem like it's hugely better than the picture they put in the article.

dsl 7y ago
Spy satellites aren't like camera phones. The killer feature isn't always higher resolution or being able to zoom in on a licence plate. Imagine having a big enough sensor and lens array to be able to see every individual boat in the Strait of Hormuz and track them all in real time. Or being able to pinpoint a spot on the planet and over the course of multiple passes by different birds, be able to construct a photo realistic 3D model of an oil refinery for a SEAL team to walk through in VR.
- ThisIsTheWay 7y ago
  +1. The GEOINT and IMINT communities are investing heavily in multi-source collocation and automated analysis. In the past the focus was to improve resolution, now the goal is to improve the analysis of the images with additional data. Short wave infrared, synthetic aperture radar, creation and maintenance of digital elevation models, and frequent revisit rates to aide in computer/AI monitored change detection are all high priority. Great reading here: https://www.theverge.com/2019/7/31/20746926/sentient-national-reconnaissance-office-spy-satellites-artificial-intelligence-ai https://www.theverge.com/2019/7/31/20746926/sentient-nationa...
- RyJones 7y ago
  I attended Hack The Machine[0] in Seattle a few years ago. The amount of data the had for people to work on was amazing - the Navy provided a set of AIS data with some number of collisions in it; the challenge was to use your technique to find more collisions in the broader dataset. The guy said something like the Navy had 100% of AIS data going back a decade for us to work on, because of course they do. [0]: https://www.hackthemachine.ai/ https://www.hackthemachine.ai/
- SiempreViernes 7y ago
  the Strait of Hormuz in a single frame? That's around 100 km, not gonna be a lot of pixels per meter in that image...
  - haimez 7y ago
    Why would you think that? It seems like you’re assuming some limitation that isn’t inherent to the problem.
  - phkahler 7y ago
    Terapixel imager.
  - CalChris 7y ago
    Commercially, we have gigapixel imagers [1]. (100 km^2) / (1 billion) = 0.1 m^2 https://en.wikipedia.org/wiki/Gigapixel_image https://en.wikipedia.org/wiki/Gigapixel_image
    - ArnoVW 7y ago
      1 gigapixel means 32768 pixels per side (assuming a square image).
      - TeMPOraL 7y ago
        So you now have 3 pixels per meter in each direction of the 10km x 10km square.
    - mkl 6y ago
      100 km^2 ≠ (100 km)^2. For a 100km by 100km image, 1 gigapixel gives you pixels 3.16m by 3.16m, 10m^2 each. Probably still enough to track boats though.
cdash 7y ago
We saw a capability, we don't have a clue if that is the best capability available.

dsl 7y ago

Spy satellites aren't like camera phones. The killer feature isn't always higher resolution or being able to zoom in on a licence plate. Imagine having a big enough sensor and lens array to be able to see every individual boat in the Strait of Hormuz and track them all in real time. Or being able to pinpoint a spot on the planet and over the course of multiple passes by different birds, be able to construct a photo realistic 3D model of an oil refinery for a SEAL team to walk through in VR.

ThisIsTheWay 7y ago
+1. The GEOINT and IMINT communities are investing heavily in multi-source collocation and automated analysis. In the past the focus was to improve resolution, now the goal is to improve the analysis of the images with additional data. Short wave infrared, synthetic aperture radar, creation and maintenance of digital elevation models, and frequent revisit rates to aide in computer/AI monitored change detection are all high priority. Great reading here: https://www.theverge.com/2019/7/31/20746926/sentient-national-reconnaissance-office-spy-satellites-artificial-intelligence-ai https://www.theverge.com/2019/7/31/20746926/sentient-nationa...
RyJones 7y ago
I attended Hack The Machine[0] in Seattle a few years ago. The amount of data the had for people to work on was amazing - the Navy provided a set of AIS data with some number of collisions in it; the challenge was to use your technique to find more collisions in the broader dataset. The guy said something like the Navy had 100% of AIS data going back a decade for us to work on, because of course they do. [0]: https://www.hackthemachine.ai/ https://www.hackthemachine.ai/
SiempreViernes 7y ago
the Strait of Hormuz in a single frame? That's around 100 km, not gonna be a lot of pixels per meter in that image...
- haimez 7y ago
  Why would you think that? It seems like you’re assuming some limitation that isn’t inherent to the problem.
- phkahler 7y ago
  Terapixel imager.
- CalChris 7y ago
  Commercially, we have gigapixel imagers [1]. (100 km^2) / (1 billion) = 0.1 m^2 https://en.wikipedia.org/wiki/Gigapixel_image https://en.wikipedia.org/wiki/Gigapixel_image
  - ArnoVW 7y ago
    1 gigapixel means 32768 pixels per side (assuming a square image).
    - TeMPOraL 7y ago
      So you now have 3 pixels per meter in each direction of the 10km x 10km square.
  - mkl 6y ago
    100 km^2 ≠ (100 km)^2. For a 100km by 100km image, 1 gigapixel gives you pixels 3.16m by 3.16m, 10m^2 each. Probably still enough to track boats though.

cdash 7y ago

We saw a capability, we don't have a clue if that is the best capability available.