Schneier on Security
A blog covering security and security technology.
« WikiLeaks |
| Security Trade-Offs in Crayfish »
June 24, 2010
TacSat-3 "Hyperspectral" Spy Satellite
The idea of hyperspectral sensing is not, however, merely to "see" in the usual sense of optical telescopes, infrared nightscopes and/or thermal imagers. This kind of detection is used on spy satellites and other surveillance systems, but it suffers from the so-called "drinking straw effect" -- that is, you can only view a small area in enough detail to pick out information of interest. It's impossible to cover an entire nation or region in any length of time by such means; you have to know where to look in advance.
Hyperspectral imaging works differently. It's based on the same principle as the spectrometry used in astronomy and other scientific fields - that some classes of objects and substances will emit a unique set of wavelengths when stimulated by energy. In this case, everything on the surface below the satellite is being stimulated by sunlight to emit its unique spectral fingerprint.
By scanning across a wide spectrum all at once across a wide area, it's then possible to use a powerful computer to crunch through all wavelengths coming from all points on the surface below (the so-called "hyperspectral cube", made up of the full spectrum coming from all points on a two-dimensional surface).
If the sensor is good enough and the computer crunching powerful and discriminating enough, the satellite can then identify a set of points on the surface where substances or objects of interest are to be found, and supply map coordinates for these. This is a tiny amount of data compared to the original "hyperspectral cube" generated by ARTEMIS and crunched by the satellite's onboard processors, and as such it can be downloaded to a portable ground terminal (rather than a one with a big high-bandwidth dish). Within ten minutes of the TacSat passing overhead, laptop-sized ROVER ground terminals can be marking points of interest on a map for combat troops nearby.
Posted on June 24, 2010 at 1:21 PM
• 26 Comments
To receive these entries once a month by e-mail, sign up for the Crypto-Gram Newsletter.
Looking for 4 Dimensional Ghosts? If Love Hewitt is the PM...sign me up!
Jennifer is mine!
Hmm Jenifer Love Hewitt...
You know wat they say about beauty being only skin deep well to a similar extent this " hyperspectral sensing" is only skin deep...
All technology has it's limitations and this ones probably a couple of rolls of Baco foil and a tarp with solid objects...
Oddly though similar technology can see smells so your meth lab etc may well be easily seen as would certain crops.
And this may well be it's strong point in locating hidden manufacturing by it's effluent put out into the environment.
"In this case, everything on the surface below the satellite is being stimulated by sunlight to emit its unique spectral fingerprint."
Its color? And temperature?
Sounds snakey to me, but I'll bet the military will spend plenty for it.
I like the idea of using hyperspectral imaging to find risks in agriculture, roads and other infrastructure.
This combat application however reminds me of the effect of inflatable tanks and trucks in WWII.
I'm certain there already is way to issue bogus "unique set of wavelengths" to force false positives. It also is likely possible to hide, mask and distort wavelengths as Popular Science suggested in 2001.
Flight paths of the sensors are predictable and apparently they do not work through clouds or at night.
I guess I should also say that when I mean "hide" I think of how the Eritrean and Tigraen rebels maintained invisibility despite technology and manpower of the 300,000 standing Soviet-backed Ethiopian army in the late 1980s.
I'm wondering how this would work in an urban environment. We often think of concealment and guerrilla activity being conducted in wilderness, but it doesn't have to be as the Israeli's learned in Lebanon. I think there would tend to be a lot more false positives in an urban environment.
For those interested in the background to Space Imaging of Earth there is a nice little tutorial that the Center for Remote Imaging CRISP have up on their website.
This page has info on what it's all about pluss some nice pictures (look at the IKONOS 1m image showing vehicles at a junction)
There is by the way another problem with Hyperspectral imaging which comms engineers refer to as the 'link budget'.
If you assume that each "Pixel" in the image stack has one hundred hyperspectral layers at eleven bits each and each pixel represents a 0.25m square on the earths surface and the instantanious view is upto 100Km long that is one heck of a lot of raw data thats 53MByte of raw data for a strip. Now if you assume the earth is aproximatly 32,000Km around and an orbit takes ~90mins you have something like 30,000 of these image strips a second which gives a raw data rate of around 1.5TByte/Sec which is probably going to double with framing and error correction etc.
So the raw data rate is way to high to send via a radio link and definatly pushing your luck via a cutting edge optical link in near perfect conditions.
So it's a safe bet most of the processing of raw data (in real time) is done up in the satellite.
Which also means there is a high probability it can only see what it is told to look for at any point in time so it is going to miss a lot of information...
Oh and there is another problem due to atmospheric absorbtion and scatering and you could call it "spectral gateing". Basically the Earths atmosphere is only transparent at a limited number of points in the spectrum and translucent at best at most other points. Sadly most of the transparent gates don't coincide nicely with spectral points you would want to use. Which means you have to synthesize the points from the adjacent gate points (which by the way is how humans see colour). Which means that the system has distinct limitations and can be spoofed if enough info is known about which spectral gate points are used...
Spectral imaging is not new or vaporware. See the Landsat program: http://landsat.gsfc.nasa.gov/about/index.html
Also, it is used to analyze the composition of stars, nebulae, dust clouds, etc. in astronomy.
What's cool here is the hyper spectral and wide field of view and computerized analysis aspects.
Of course it is not a complete solution and can be spoofed, but it should open a new window of intel on the planet.
I don't see how hyperspatial imaging doesn't suffer from the same drinking straw effect. But worse. Surely the more bands you monitor, the more data you need.
The only way around the drinking straw effect is to do more image processing, and more automated feature detection. But, I guess more wavebands improves the possibility for feature detection.
This sounds like an extension of the multiple wavelength image distortion analysis techniques that have been used for spying from afar for decades now. I.e. nothing ground-breaking, just an incremental advance in a longstanding technology.
Wow... Maybe 1.5TB/sec of data (which it may or may not really be able to process and which it may or may not entirely pick up), but according to page 7 of http://www.responsivespace.com/Papers/... it has a 274Mb/sec downlink. Not too shabby and certainly able to pass lots of local information...
@ David t-g
"... it has a 274Mb/sec downlink. Not too shabby and certainly able to pass lots of loca information.."
That is a maximum more normaly you would be looking at 10's K/s for ROVER stations.
Since my earlier napkin estimate I've been digging around.
It turns out the ARTIMS package is based on Comercial Of The Shelf (COTS) components...
It's charexteristics are therefore fairly well known, and NASA and ESA actually have fielded better equipment.
Simply it consists of a comercial teloscope and co located spectrometer and a high res monocrome image camera. The spectrometer is only 10bit and is far from state of the art.
Apparently the software runs on a G4 industrial controler (similar to PC104) and is primarily designed to output detected targets as text messages not imagery.
Although this is not something you could knock up in your average garage, it is well within the capability of the likes of Israel who poped their fith spy satellite up just a few days ago.
The cost of the device would also be well within the budgets of many many countries and many corperations.
For instance the French SPOT could pop one of these up as could one or two other comercial Earth imaging organisations.
Can't remember which movie it was, where the meeting is taking place in the desert and the first thing that happens is those monstruous Yukon trucks going around in a circle.
Thick clouds of dust rise into the air and then the tactical advantage of the satellite goes bust...
@ Louis: typical movie plot and not particularly wise for the meeting parties. Trucks normally don't run in circles so it's a potential guided missile target :-)
They'd better meet during a natural sandstorm - or simply inside a building, cave or other structure...
The last paragraph explains that the images are processed inside the satellite and it only sends the coordinates of interesting areas to the ground terminal.
So there's no bandwidth issue. They just need large memory space and fast chips inside the satellite to process all this data.
@Peter A. "They'd better meet during a natural sandstorm - or simply inside a building, cave or other structure"
Yep you're right. Buildings and caves are easier to predict.
I hope it can help to protect our troops from those IEDs.
For an idea of specs I would look at the M^3 instrument flown by NASA on the Chandrayaan-1 spacecraft. It's basically the same technology:
Great way to scan for pot farms....
Illegally of course, but that hasn't been stopping our government for quite some time now.
@DCFusor "Illegally of course"
We may not like the decisions being made but Remote Sensing anywhere outsdie the home is legal and been upheld. A helicopter identifying pot plants in a back yard does not violate homeowner soverenity. Scanning national forests for farms or farmers acreage for hidden crops is legal.
If it senses within the domocile it's a different case.
> Its color? And temperature?
> Sounds snakey to me, but I'll bet the military will spend plenty for it.
Sure, its colour is what you get. But to the human eye -- or any other three channel colour sensor -- green is green. To spectral analysis, there are an infinite number of different ways of composing the colour sensation of "green" from individual spectral lines.
Some of those "greens" are reflected by, say, chlorophyll in plants. Others are reflected by, say, camouflage paint. To the naked eye, and to a conventional camera, they may appear identical, but under hyperspectral imagery (HSI) the difference is obvious; the green-painted camouflaged missile launcher sticks out as plainly as it would to the naked eye if it was painted blaze orange.
The data link required to actually download a full spectrum from every pixel, in real time, is staggering, but one of the cute features of HSI is that often you don't need to. You upload spectra of interest to the on-board processor, and tell it "whilst scanning this desert region, any time you spot a bit of Toyota Paint Code 416 ('Sand dune beige'), let us know about it and we'll take a closer look."
This is not snake oil; it has been used in geological and agricultural remote sensing applications for years, and has recently started to be used in search-and-rescue services. There are even low cost hobbyist builds (technically "multispectral" rather than "hyperspectral", since they have more then 3 colour channels but less than the dozens to hundreds considered to constitute true HSI.) What is probably novel about TacSat-3 -- if it is novel, that is! -- is developing enough on-board processing power to handle HSI on a platform that can't land and hand over a massive hard drive every day. This is harder than it sounds, because satellites can only use radiation hardened processors, which are usually a couple of generations behind the ground-based state of the art.
That's why they send up a decommissioned Cobra around here. With FLIR, the stuff stands out like a sore thumb, and when the weed quits being grown in the open, they just start looking for hot buildings and berms...
"I hope it can help to protect our troops from those IEDs"
For a couple of reasons, the first being a bomb under a real rock looks just like a real rock. So with care of placing small to medium IED's would not betray their presence. Secondly this satellite is in Low Earth Orbit (LEO) around 500Km up doing around 22,000Km/hour, with a primary mirror of small size. The realistic resolving size is probably going to be in meters not fractions of a meter so anything smaller than a car is not realy going to show up due to uncertainty and signal to signal ratio down near the equivalent of the pixilation noise floor. There are signal processing techniques that might improve things but realisticaly an IED the size of a medium sized suitcase with carefully sifted dirt or rocks over the top is not going to be seen.
"I would look at the M^3 instrument flown by NASA on the Chandrayaan- spacecraft. It's basically the same technology"
The optical front end on TacSat-3 appears to be not quite as good as it's all COTS gear (so yes you could build this yourself if you had the money). However I did dig up a side refrence that indicted NASA "tuned it".
"The data link required to actually download a ful spectrum from every pixel, in real time, is staggering, but one of the cute features of HSI is that often you don't need to. You upload spectr of interest to the on-board processor, and tell it "whilst scanning this desert region, any time you spot a bit of Toyota Paint Code 416 ('Sand dune beige'), let us know about it and we'll take a closer look."
And this is the reason it's a Tac(tical) Satellite not an Intel Satellite, and it's also one of it's main weaknesses. You have to tell it exactly what to look for for it to see it. Thus if you told it to look for mouse fur it would miss the herd of stampeading elephants heading right for your position.
There is a reason the likes of the NRO and NSA download and store everything they can. They can roll the clock back and find what they missed and thus potentialy learn from previous failings (which is what Intel realy is all about these days).
Put simply "you have to know what to look for so you can spot it before it takes a bite out of you".
Just noticed a little mistake in my above...
Assume Earth is 25,000 miles in circumfrance or 40,000Km therefore with the diameter to the LEO hight being another 1000Km and 90mins orbit you would have 43142 * 6/9 = 28,761Km/h
Moral don't do maths in your head before your breki coffee ;)
How much of this is hyperspectral hype and how much of it is an objective advance in surveillance technology? The article linked reads like a reformatted PR hand-out. What can it find--i.e. what sorts of objects emit a spectrum that the sat can identify at least in theory? What sorts of false positives does the sat get? What sorts of false negatives? Are there countermeasures (i.e. electronic and/or physical camouflage)? 'Spoofing' seems to suggest 'false positive'. One would of course want false negatives for his gear.
Schneier.com is a personal website. Opinions expressed are not necessarily those of Co3 Systems, Inc.