Spectrometer Introduction, Tear-down, and Data Analysis for Plant Phenotyping
Vložit
- čas přidán 22. 10. 2017
- Spectrometers decompose light into a rainbow of colors and then measure each color accurately. In this video I tear-down a spectrometer, describe the theory, and use of a spectrometer. A recent leaf clip spectrometer is used to for a plant phenotyping application to predict tree lead species.
Channel Love: Likes are free. paypal.me/KentVanderVelden
Business Inquiries, I'm a freelance scientific programmer: kent@confluencerd.com
Finally, I demo software that I've written for storing, organizing, analyzing (using PLS-DA), and reporting spectra. If you need help with a plant phenotyping project, let me know, I may be able to help your project.
The geranium drought experiment is not fully described due to time. There's additional background information that's needed for understand water absorption spectra. The device used in the video is not ideal for a drought application and so the results are not clear. Let me know if you would more details.
The spectrometer that is tore-down is from Ocean Optics. The leaf spectrometer is from CID Bioscience. A Microsoft Surface tablet is used to power the spectrometer and collect data.
CID CI-710 Leaf Spectrometer: www.cid-inc.com/plant-science...
Ocean Optics USB4000: oceanoptics.com/product/usb40...
PLS-DA paper ("So you think you can PLS-DA," Ruiz-Perez and Narasimhan): www.biorxiv.org/content/biorx...
Other PLS resources:
Properties of Partial Least Squares (PLS) Regression, and differences between Algorithms: www.eigenvector.com/Docs/Wise_...
A survey of Partial Least Squares (PLS) Methods, with Emphasis on the Two-Block Case: www.stat.washington.edu/resea...
Diffraction grating explanation:
• Diffraction grating | ...
• Double & Single Slit E...
#KvvCreates, #Science, #Spectroscopy - Věda a technologie
Check out my most recent video: goo.gl/Jj7cU1
Tesla500 (creator of the Chronos camera) has a great video with many examples using a similar spectrometer: czcams.com/video/7rXAP-voWfI/video.html
Thank you for all the great comments, I hope you have a wonderful day!
Loved it, there's a ton of science in this video.
Well done :)
Thank you! In case you'll interested, I just put up some tools and results of using a FLIR thermo camera in my GitHub for plant phenotyping (and measuring heat sinking of a servo motor on a millig machine)
very useful video thanks
🌱🧪🌈I found this fascinating. I think i’ll try the kitchen pigment spectroscopy thingy for some gardening fun. Thank you for giving me food for thought.
Awesome! It's fun to see that colors are hidden from view, and nothing too fancy is necessary. I build equipment in this area for scientists, but would like to make something for the home or urban gardener or small farmer.
Thank You
Wow. Thanks for the awesome explanation!
I'm currently gathering some information about spectrometers in order to build my own in the next couple of months maybe. :)
Greetings from Germany, keep going on!
Thank you for commenting. Comments really do help a lot. You've probably already seen, but if you want to make a spectrometer similar to the Ocean Optics one that I showed, and depending on how much you want to do from scratch, each piece is generally available on eBay, including the optics bench, slit, grating, mirrors, linear CCD, PCB, etc. Have you seen TI's DLP based spectrometer? It's a little different in design, www.ti.com/lit/ml/dlpb006/dlpb006.pdf Another cool design for imaging are acousto-optic tunable filters (AOTFs). Please let me know how your build goes. Best wishes, Kent
Didn't know the DLP type, but looks pretty interesting!
Parts are a bit complicated to get here in Germany on ebay, but i've shot at least a diffraction mirror, the slit should be the least problem (razorblades should do the job).
My Plan is to make a hyperspectral camera, where the intensity and wavelength "calibrating" isn't that critical. But it would be pretty interesting to see everyday objects in different parts of the spectrum. :)
At my last job I worked on hyperspectral cameras and data from them. Fascinating tools! You are building a pushbroom style? Please stay in touch. I'd like to follow your progress. I want to build a hyperspectral camera as well.
What did you work on with such a camera?
Hyperspectral data can be represented as 3D-Images to my knowledge - so to capture a hyperspectral image I'd need a 3D spectral sensor. Which is pretty much impossible with todays technology. (2D + spectral information)
I want to use a true monochrome CMOS in combination with a slit and a diffraction grating to capture one line, where the y direction of the sensor represents the spectrum (from 400-800nm roughly) of one pixel. To capture an image a galvo mirror pans the optical path. One complete picture would roughly take half a minute at decend full-frame frame rates. Should be possible, shouldn't it?
Absolutely is possible. If simpler, a mirror on a stepper motor or a fixed angle motor with the entire camera moved on rails (like a flatbed scanner) will work. Stepper motors are often preferred in optics because they don't dither. Even UAV mounting is possible with the UAV forward motion as the scan. This configuration is called pushbroom, with the slit like a broom, moving forward, collecting pixels. The duration is largely determined by the intensity of the light and distance of the light-subject-camera path. I was on a team using various camera technologies for measuring plant health. One camera we used was from Headwall Photonics. Best of success to you.
great informational video love from india punjabi🥰
Hey Kent! Thanks for the video! I'm curious how much time it took to do the plant leaf recognition research (including collecting samples, creating the software, etc). And was it related to a current work project, or purely for fun? I'm always blown away with how much you are able to accomplish. Good luck on the home-made spectrometer project! That will be an interesting video to watch :-)
Thank you Nathan. A few months are into this project (for a client) and I only showed screens and examples that are generic. Everything is easy once one has the experience, and I have some experience with this sort of instrumentation. Identifying the hardware and working with vendors takes time. Collecting samples is fast. Driving to and walking the Arboretum took most of the sampling time. Scientific software generally requires a lot of prototyping to find the right method. (Jupyter is a good environment for this.) So while the final implementation can be fast, the total time spent on the software based research is easily the longest. In this case, many weeks. However, now that that work is done, I can use the components in larger projects. Reuse of hardware and software (and experience.) The reuse is where one earns a reasonable amount. The first time one barely breaks even. I skipped the droughted geranium results because additional background would be needed, and the expected signal at 970nm, the 2nd overtone water absorption band, was not clear. Hope that helps. Best wishes, Kent
Thank you for the video - now I know a bit more :) I think a portable color sensor like Palette Cube or Nix Pro Color Sensor would be an awesome project. Especially with opensource software
Hi Dmitry, thank you. Those appear to be colorimeters, based on price. I don't know how those work, though they generate tristimulus values instead of a full spectrum. I assume multiple phototranistors with different filters and software for deconvolution? It's quite possible they would be sufficient for plants. Do PCA of a plant's visible spectrum and the first three PCs capture most of the data. Generally with plants one is primarily comparing "greenness" as a proxy for health or regressing spectra against chemical measurements like nitrogen. There's a lot of noise in plant data too (measurement error, dust, water, spray residue, etc.), and so high resolution measurements may be overkill. Colorimeters are unlikely to work for emissions spectrum measurements or identifying materials. You have a great suggestion. Maybe start with a colorimeter and reverse engineer the protocol so it's useful in other applications. I have a Spyder I could try that on. I was thinking of making the parts available for a decent teaching spectrometer for those willing to do some assembly and calibration, maybe STEM programs. Definitely open source. There are a few designs on Hack-a-day and bits and pieces available on eBay. Most of the electronics used in the decent spectrometers is pretty old tech and software is limiting. The prices for new diffraction grating based spectrometers is 50x of the Palette Cube, so outside the range of what is reasonable for primary schools (NIR spectrometers are worse.) A lot of that cost is quickly diminishing returns, in my opinion. There are inexpensive transmission diffraction gratings based spectrometers, such as Ocean Optics's Spark. A really interesting one that uses a different method is TI's DLP based spectrometer. Long response, but I find this stuff pretty interesting.
Great,
Would mind if I ask, I have interest to know the part number of the CCD in that device.
I have been trying to make choice on CCD sensor for a project. This may be a good start point.
Nice one Kent. Wish I was as smart as you. Keep on posting buddy ;-). Cheers, Aaron.
Thank you very much Aaron, that means a ton. But honestly, I know a lot about very few things :) And, when the electricity goes out, I'll be completely useless... For work, I'm helping to improve plant measurements and data analytics for others. They have amazing goals such as growing crops symbiotically instead nuking with synthetic chemicals. Mine is only a support role, but I love the vision, and I'm glad to be helping. Best wishes from the getting-colder-everyday Midwest.
Are you using Python with Scikit-Learn library for the machine learning?
And Python + Flask for the web application?
It's an amazing video, great job!
Hi Knet.
Greetings from Brazil. Great video by the way.
Any tips for the construction of one of these spectrometers. I already work with deep learning for precision agriculture. But would like to use spectral images for determination of plant nutrition etc.
Best regards
Hi Luiz, there are a few folks who have posted there designs on CZcams. There's a lot I don't know or understand so best to consult an optics engineer or similar text. Going to be hard to match quality of purchased unit without a lot of cost. We work on similar fields. See www.confluencerd.com for examples of what I'm making for plant phenotyping. Contact me by email if interested in sharing more. Best wishes, Kent
thank you for this clean explanation , i'm working these days on a project of detecting chemical decomposition of fodder samples using NIR spectroscopy, and i'm planning to follow the same path as your project. I'm interesting of having more information about the softwar part, would you have some ressources that can help me understand the PLS process? knowing that it will be my first time working with such tool.
Sounds like a fantastic project! I grew up on farms and loved the smell of silage. You're asking a tough question to answer not knowing where you're coming from. If you have a chemistry background you'll look at chemometrics vs. a mathematics or statistics approach. That's for the theory, but you may actually want to apply immediately and then it matters if you're using Matlab, Python, etc. Are you after broad crude fat, moisture, protein, and fiber measurements or specific species?
Maybe start with
www.scinapse.io/papers/2158863190#fullText
for the theory? Look for papers and patents where others have done similar work that have methods written well enough to be tutorials and replicate the paper. Just some ideas that come to mind. Wishing you the best of success!
Nice video. Do you do preprocessing on spectra before applying partial least squares?
Thank you! To keep this video consistent, I only enabled feature scaling, but I implemented methods that USDA-UNL published on preprocessing spectra for PLS and PCA. I like their approach because they encourage visually exploring varying levels of preprocessing to identify features that sensitive to preprocessing. It's not so easy to put that into an automated system, but none of this stuff really is. If you have suggestions, let me know. I'm leaning more to training people to use a bit of Python, modify parameters and rerun, than to try to do everything on a webpage. Saves me time (and danger of implementing code that's never used) and gives them flexibility ("teach a man to fish.")
Great video Kent! Thanks for posting. I am interested in deuterium depleted water. Deuterium is an isotope of hydrogen, witch could be causing ill heath effects. I am trying to vacuum distill my water to separate H2O and D2O. What is the best or how can a spectrometer test my water?
This is well outside of my specialty. Out of curiosity, I did some brief reading and the links are below. The first two might be good for comparing energy and efficiency of generation methods: generating light water directly or as a by-product of generating heavy-water. The paper on light potable water generation has their testing method for in the Analysis of Deuterium Content section, which indicates inline gas chromatography. The last paper uses IR spectroscopy. The convolution between the two species, high ratio of H2O to D2O concentration, and cost of the required spectrometer may make the process impractical. Instead of testing regularly, would you be satisfied implementing the potable light water system and trusting their numbers or perhaps testing occasionally at a mail-order lab? You have an interesting project that requires the advice of a specialist, perhaps several. Best of success to you! Stay healthy!
cns-snc.ca/media/Bulletin/A_Miller_Heavy_Water.pdf
sci-hub.tw/doi.org/10.1021/ie101820f
sci-hub.tw/doi.org/10.1016/0020-0891(70)90003-5
is there a way to get the 1 pixel array sensor? I would love to build my own spectrometer
Hi from Ukraine! Great video Thank you
That's amazing! Thank you Aleksandr for saying Hi.
Another great video, really liked when you dunked the paper. By the way, do you ever sleep?
Thank you Paul! Shave an hour of sleep each day, then over a month, an entire day is earned back. Actually, I love to sleep, it's one of the top three "nutrients." :)
They sell budget macro lenses for cellphones, you could sell a budget diffraction grating (hologram film) mounted to such a macrolens holder. You software could take a photo of the leaf shape and a rudimentary spectrogram of the colour to increase your 55% to closer to 100% 🙂
Can these spectrometers distinguish various types of gases in the air using sunlight? If not, what would be a setup that can do this? I am interested in methane gas.
Hi Riz, you may want to look at MWIR spectrometers given the absorption spectrum of methane. Check on atmosperic absorption in this range. Going to need to calibrate and a light source other than the sun may need easier. I've only worked with spectrometers in the visible to NIR range, like the one in the video. So hopefully I've not misspoken. You have a very cool sounding project. Please let me know if I can be of any help.
www.meteor.iastate.edu/gccourse/forcing/images/image7.gif
Hello Kent, our company needs to measure betanin content in our byproduct - red beet peels waste. Our university offers UHPC/MS method, which is more expensive , but as i understand very precise and also there is Betalain( as i understand which is 90percent betanin) concentration measure with UV-Vis spectroscopy method which is much cheaper, but not that accurate. The scale of project is quite vast and intense. Could you provide some insight about these methods and which one you advise to go with?
Hi Julius, thank you for asking, I could likely help with some aspect. The accuracy will probably be limited by convolution with signals from overlapping pigments. Please send me an email with any other details you can share, and let's plan on talking. Have great day!
Also, you can find more contact information on www.confluencerd.com/ Would be glad to help. Have a great day!
Hi Sir, It was a very nice video. I would like to know, how to build a spectrometer that works in the wavelength range of 1600 nm - 6000 nm?
From the video, I think that we need to choose a proper diffraction grating that can split the light and the corresponding detector. Would you please suggest what is the diffraction grating that I need to purchase and where can I get it.
I don't have experience with such a wide range. You may want to start with the selection of the detector. Multiple detectors may be needed to cover this wide of range.
Also, diffraction gratings are one type of dispersive element, but there are others. Maybe a different one will be more available?
@@kentvandervelden Thank you sir.
Thanks Kent, now I feel dumb.
Hello sir, how to make starting wavelenght from 200nm? by Grating angle?
Yes, in general, I believe. In the one I showed, UV probably spilled off the side of the sensor. You'd need a sensor that is sensitive to UV, but silicon sensors can be found in the 200-1100nm range. Check their spec. Thank you
@@kentvandervelden 🧡sincerely appreatiate your kind reply. Valuable information for me!
Интересно а биохимию потянет такой спектрометр?
NIR spectroscopy would be better for compound detection, but overall plant health is a good application of VIS spectroscopy.
55% accuracy is really bad...
I wondered how specific the UV/Vis (reflection)spectra for different types of leaves is.
Thanks for asking, it's pretty interesting. That's a measure of classifying specific species in a diverse species population. I'd have to review the results, but from memory, and it's been a long time, is that certain groups are easily discernible while others are not. 11/19, or something like that. Look at the confusion matrix. To differentiate between species not discernible with spectra alone, adding a small amount of morphology should clinch it. Like you said, there is not be a lot of difference in visible spectra between species. Just a difference in ratio of a few pigments. Some species are easy, and can be seen with your eye, others are apparently not. There are a few handheld spectrometer that interfaces to smartphones and uses crowd sourced data. They make pretty great claims, but the reviews are so-so probably for similar reasons. LinkSquare and SCiO are a couple. Peace
www.photonics.com/Articles/Smartphone_Spectroscopy_Promises_a_Data-Rich/a63594
www.cnet.com/reviews/stratio-linksquare-preview/
spectrum.ieee.org/view-from-the-valley/at-work/start-ups/israeli-startup-consumer-physics-says-its-scio-food-analyzer-is-finally-ready-for-prime-timeso-we-took-it-grocery-shopping
If interested in trees in particular, check out "Review of studies on tree species classification from remotely sensed data" F.E. Fassnacht et al. 2016. They include results from studies using similar and much fancier spectrometers as well as other sensor types. While prepping a UAV remote sensing review I read this paper and thought of you. Amazing coincidence
Is not running
Sorry, what's not running?
@@kentvandervelden
Hey.
At 0;47 the video stops here in my Mex place.
Do not know why.
Its been each single time I want to watch it.
Argh, we're at the mercy of CZcams :( Thanks for mentioning it
@@kentvandervelden
Your welcome.
Its a pitty I cannot watch it.
Cheers.
If you get traction on these types of videos, lets make some more of them! I know of a bottomless garbage can...
Hi Jacob, thank you, I still have the last care package from you to put to use. I remember you saying: "I know how these things go, in a year when you get to it... " Well, we are at 14 months :) There has been some consulting inquiries that have come from the spectrometer video. I would like to show more of the other work that I do, the process of learning and implementing. Recently I've been implementing image segmentation with convolution neural networks and the results are fantastic in terms of simplifying the image processing routines. I have yet to find a good way to describing that sort of work here, without impeding the gig, but it's what needs to be done to find more like minded people, consulting gigs, etc. Long tangent shortened, you have a good idea of what might be interesting, so if you see something that's going into the garbage, do the environmentally conscious thing, please grab it. :) Stay healthy!
Hola
Hello! :)
Kent VanderVelden hello i'm i'm a escool
Pleasure to meet you, I'm from the corn belt of the USA. Are you in Catalonia? I was reading about Andorra recently, which is interestingly positioned on the map.
Help me please i'm i'm China Hong Kong nooooooooo
You are too advanced for me a beginner
This makes me sad to hear as it means I did a poor job at explaining. Sorry, I wish I would have done better.