Some of you may remember the SCiO, originally a Kickstarter darling back in 2014 that promised people a pocket-sized micro spectrometer. It was claimed to be able to scan and determine the composition of everything from fruits and produce to your own body. The road from successful crowdsourcing to production was uncertain and never free from skepticism regarding the promised capabilities, but the folks at [Sparkfun] obtained a unit and promptly decided to tear it down to see what was inside, and share what they found.
The main feature inside the SCiO is the optical sensor, which consists of a custom-made NIR spectrometer. By analyzing the different wavelengths that reflect off an object, the unit can make judgments about what the object is made of. The SCiO was clearly never built to be disassembled, but [Sparkfun] pulls everything apart and provides some interesting photos of a custom-made optical unit with an array of different sensors, various filters, apertures, and a microlens array.
It’s pretty interesting to see inside the SCiO’s hardware, which unfortunately required destructive disassembly of the unit in question. The basic concept of portable spectroscopy is solid, as shown by projects such as the Farmcorder which is intended to measure plant health, and the DIY USB spectrometer which uses a webcam as the sensor.
“The basic concept of portable spectroscopy is solid…”
No doubt, but I assume there are some limitations as well.
When this was out, i remember digging through some research literature to see. My understanding is that it can be reasonably precise to identify one of few predetermined things, but the moment when many things come into play, the error rate is high.
I’ve played with the SCiO and looked at the raw output data. Based on repeated trials of testing of different materials, it looks like you get about 10 bits of reproducible information from a measurement (10-15nm resolution over 150-200nm* with 0-6dB SNR per bin across many trials of angle and lighting). That’s what a classifier algorithm gets to work with. If the samples are diverse enough, you might classify between a few dozen of them, possibly even regressing coarsely on fat and protein content (which is known to absorb in NIR). However, as information goes, it’s pretty low. You’d get more from a single pixel of a color image. Even from visible color, you should be able to classify among an apple, pear and lemon, and map to nutritional information in a database (or have it learned directly by a nonlinear classifier).
Have others experienced similar results?
*The absolute measurement range is 740nm-900nm (they claim 1070nm, but above 900nm the sensitivity drops off almost completely).
The information bits are based on a calculation of correlation coefficient of about 0.85 at each wavelength, measured over multiple many angles lighting, etc.. per bin (where # bins is estimated to be (900nm-740nm)/15nm (resolution). This gives a total mutual information of, 160/15*log2(sqrt(1/(1-0.852 ))) = 9.86bits
This assumes the signal of interest and the measurement noise are Gaussian distributed, which appears approximately right.
For comparison, something like this http://oceanoptics.com/product/nirquest512/ gives 2081 bits per measurement using the same calculation..
This illustrates the limitations (and issues) of any first gen device, as well as trying to market something that is a good idea, but just not quite ready to be sold.
Looking at the hardware design, they were probably trying a bit too hard to forcefully marry the electronics into the marketing driven design of the case.
Putting the lamp and sensor on a fine pitch multilayer rigid-flex board and having to wire-bond the sensor array connections to the rigid part of that board? Looks like a pretty expensive piece to fabricate…
I believe the Scio would have been way more successful if it had allowed for raw access to the data. I would still try to procure one if raw data were available for $300 (+ perhaps some hacking). As it stands the raw spectrum isn’t even available when paying for the 950$ Version “Development Toolkit – SCiO Sensor + Development License” (see https://www.consumerphysics.com/order-scio/)
Anyway, I recently held off on purchasing a SCiO for teardown because:
1.
For those interested in the Scio internals- quite a lot of information could be gleaned from the following documents:
http://www.systemplus.fr/wp-content/uploads/2017/03/SP17304_SCIO_Spectrometer_Sample_System_Plus_Consulting.pdf
http://www.systemplus.fr/wp-content/uploads/2017/02/SP17304_SCIO_Spectrometer_Flyer_System_Plus_Consulting.pdf
2.
The following devices (which offer raw data output) seem just as interesting for people looking to experiment right away:
a) 999$ TIDA-00554 DLP (http://www.ti.com/tool/TIDA-00554)
b) 2499$ Raspberry Pi based Full NeoSpectra Micro Dev Kit (http://www.neospectra.com/shop-products/) and possibly the NeoSpectra Modules
and perhaps
c) http://www.solidscanner.com/produkte/solid-scanner-nir-scanner-kunststoffe/
I am sure there are more I haven’t listed, please post any in the comments.
-regards
Or just get a Hamamatsu: http://www.hamamatsu.com/us/en/product/category/5001/4016/index.html#measurement_light=1
There’s a breakout board with sensor I’m eyeballing on Tindie($300). But the range is UV to NIR:
https://www.tindie.com/products/PureEngineering/c12666ma-micro-spectrometer/?pt=ac_prod_search
Holey moley this is some cool stuff! I teach at a small college and have been trying to get “simple” projects for an instrumental chemistry class to make spectrometers. I would love to hear more ideas. I will look at all these links in the meantime. This type of instrument is what I had thought of but there could be a lot more… https://www.vernier.com/products/sensors/spectrometers/
They really need to understand: their target audience wants to dig into things. The marketing seems aimed at soccer moms, but I doubt they’ll get traction there first. Raw data is just a neccessity for the sort of science/tech enthusiast who would want to play with one.
I had pegged this project as a scam and let it drop off my radar. I applaud the company for producing a product. Any differences after that fact are based on reader-conceived unrealistic product expectations. They intended to produce a product for consumer consumption that they would retain full ownership of, and that would utilize user generated data to improve its functionality. The later is where they fell short. No one will spend the time and effort required to provide the quality of data required if there isn’t a short term benefit for them.
So does it actually do anything useful?
yes, literally thousands of useful things, it is just a specialized spectrometer, nothing new, nothing controversial and used every day in a lot of different industries, from farming and food production over medical research to industrial planning and that is just for starters.
Well, according to the website the 300$ consumer version has included:
a Produce Selector app,
food analysis for meats, dairy products, fish, chocolate, fruits and vegetables,
pill identification,
and the SCiO Workshop
https://www.phone.consumerphysics.com/
Its now available in micro version, Who’s gonna crack open that one up. Spectonometer should be developed or at-least the design part would easy and than we can make our customization from it. Anyone of you guys still working with it, or gained more insight. Please share more…..
has anyone tried to hack the software (aplicattion) ? if you cannot get the raw spectra it is useless and the price to do it is bit expensive.
We developed an app that makes it more useful for field testing. You can specify multiple models, multiple scans, work offline. Results all logged to a cloud database for download and analysis. Android only at present Analyser for Scio. Still a bit beta – needs a bit of field testing.
Hi Paul.
I have just bought a consumer physics Scio IR spectrometer. I would be interested in helping you with your App development.
I have a need for some affordable IR spectrometer systems. I am a engineer /Scientist who was trained by NPL. Experience includes designing real time embedded scientific instrumention. I did instrumentation to analyse the molecular structure of polymers. Obviously this is not a laboratory grade instrument , but the extra cost involved in doing anything useful is a rip off. The basic unit is quite impressive. I considered looking at the IR spectrum & the Bluetooth signal.
Im looking to build a proper spectrometer & am in contact with various manufacturers.
Good luck
Hi Marc,
I am not Paul but I would be interested in what you have to offer. I am thinking of something along the same lines (“proper spectrometer”) and would love to hear your thoughts and ideas. How can I get in touch with you? (I tried googling you but no real luck.)
Best,
Steve
Currently (barely pre-Christmas 2018), you may order one, for personal use only, for delivery in two months, for $299 plus taxes and shipping. To me personally, this seems like a very fair and reasonable price.
Call for commercial pricing (but you know, if you have to ask…).
The non-commercial software development kit is priced at $750. This seems like a fine strategy for a company that wishes to suppress opportunity, who does not appreciate any benefit from third party development.
Many talented people might easily pay the $300 unit cost, but find better opportunities for their next incremental discretionary $750 than the SCiO software development kit. Good luck, shortsighted developers, on your long term success.
There’s a new github project trying to hack this device. Anyone interested, head over and help out if you can: https://github.com/kebasaa/SCIO-read
Are you going to read the Bluetooth comms output from the device first ? This would be the 1st step . Do you agree?