5 ms·

Did you write up any details about your microscope build? Or could you provide a good launching point for that?

dekhn 4y ago
No, I haven't, because I don't want to support users with something I'm teaching myself with (it changes quickly as I learn that I did something wrong). In short, I use aluminum extrusion on optical breadboard to build a rigid frame and then mount linear stages based on stepper motors, smooth rods and 3d printed bearings and mounts. This forms a 3-axis stage to move the sample (could be a slide, or a dish, or whatever) which I control using FluidNC (normally intended for CNC) and some hand-written Python scripts for the UI/camera interfacing. It uses industrial vision cameras. Most components are fairly inexpensive- at least compared to scientific systems- but aren't cheap for a typical hobbyist. The breadboard is >$200, the extrusion is another $100, the cameras at least $250, and my total personal time investment (converted to salary dollars) would be ludicrous. The important advantage of my design is that I control the closed loop camera/motor control stack (with machine learning inline), and I can raster scan to make large field of view stitches with the camera moving constantly and super-short light exposures. Here's an example video tracking a tardigrade (keeping its snout as close to the center as possible) I made with the previous generation: https://www.youtube.com/watch?v=LYaMFDjC1DQ https://www.youtube.com/watch?v=LYaMFDjC1DQ
- ericbarrett 4y ago
  Fascinating. What is it eating starting at ~36:00?
  - dekhn 4y ago
    Algae. It's a mixture of https://www.carolina.com/invertebrates/water-bear-tardigrade-culture-living/133960.pr https://www.carolina.com/invertebrates/water-bear-tardigrade... and https://www.carolina.com/algae/chlorella-living/152069.pr https://www.carolina.com/algae/chlorella-living/152069.pr Then I use a special tool (really just a mouth tube that has a glass tube at the end) to single tardigrade to a glass-bottom petri dish that goes into the scope. I take movies, then label ~hundreds of frames manually with "tardigrade" "algae" and "junk" then fine-tune an existing object detector with those frames. The centroid of the object detection provides the direction to move the stages to center the detected object. The other mode is fast scanning, which I described above (move the camera continuously over a large field of view, pulsing the LED quickly/periodically and dropping any empty frames). (at work I get to play with grown-up microscopes and they are another class entirely; precision machines with extraordinary capabilities and price tags to match)
    - uneoneuno 4y ago
      I used to do species population spot-check counts of the aeration basin floc thru a crappy optical scope back in the day at the municipal wastewater plant. I loved finding the tardigrades occasionally and watched them for ages.

No, I haven't, because I don't want to support users with something I'm teaching myself with (it changes quickly as I learn that I did something wrong). In short, I use aluminum extrusion on optical breadboard to build a rigid frame and then mount linear stages based on stepper motors, smooth rods and 3d printed bearings and mounts. This forms a 3-axis stage to move the sample (could be a slide, or a dish, or whatever) which I control using FluidNC (normally intended for CNC) and some hand-written Python scripts for the UI/camera interfacing. It uses industrial vision cameras. Most components are fairly inexpensive- at least compared to scientific systems- but aren't cheap for a typical hobbyist. The breadboard is >$200, the extrusion is another $100, the cameras at least $250, and my total personal time investment (converted to salary dollars) would be ludicrous. The important advantage of my design is that I control the closed loop camera/motor control stack (with machine learning inline), and I can raster scan to make large field of view stitches with the camera moving constantly and super-short light exposures. Here's an example video tracking a tardigrade (keeping its snout as close to the center as possible) I made with the previous generation: https://www.youtube.com/watch?v=LYaMFDjC1DQ https://www.youtube.com/watch?v=LYaMFDjC1DQ

ericbarrett 4y ago
Fascinating. What is it eating starting at ~36:00?
- dekhn 4y ago
  Algae. It's a mixture of https://www.carolina.com/invertebrates/water-bear-tardigrade-culture-living/133960.pr https://www.carolina.com/invertebrates/water-bear-tardigrade... and https://www.carolina.com/algae/chlorella-living/152069.pr https://www.carolina.com/algae/chlorella-living/152069.pr Then I use a special tool (really just a mouth tube that has a glass tube at the end) to single tardigrade to a glass-bottom petri dish that goes into the scope. I take movies, then label ~hundreds of frames manually with "tardigrade" "algae" and "junk" then fine-tune an existing object detector with those frames. The centroid of the object detection provides the direction to move the stages to center the detected object. The other mode is fast scanning, which I described above (move the camera continuously over a large field of view, pulsing the LED quickly/periodically and dropping any empty frames). (at work I get to play with grown-up microscopes and they are another class entirely; precision machines with extraordinary capabilities and price tags to match)
  - uneoneuno 4y ago
    I used to do species population spot-check counts of the aeration basin floc thru a crappy optical scope back in the day at the municipal wastewater plant. I loved finding the tardigrades occasionally and watched them for ages.