Automatic object recognition and automatic scale
Alex wrote 03/07/2019 at 16:22 • 1 pointI make this question seeking guidance. I want to set up an automatic insect larvae counting system (yes, something unusual), which would then be transferred to a scale. I'm contemplating several options like OpenCV and Machine Learning.
The larvae are difficult to handle: unlike nuts or screws from similar projects, the larvae move, twist, widen or slip through any slit, and often have the habit of not following the marked path. That's why I thought of something different for the count.
One option I'm considering is to have a small conveyor belt from which to "throw" the larvae on a surface. Every time a larva hit the surface, I could record this event as a larva, and thus carry out the count. The problem is that I don't find any precedents for systems of this type (that record an "impact" event as a count), or if they exist, how one could be built.
Thank you for your help and greetings.
Alex
Discussions
Become a Hackaday.io Member
Create an account to leave a comment. Already have an account? Log In.
Hi Alex - I 2nd the 'photo-interrupter' idea. I've done similar things with bees. An issue is getting them spaced out so two don't count as one big one. You may want to consider making a vibrating trough - a cell phone vibrator motor mounted on a piece of sheet metal (aluminum heat duct pipe?) bent to make a long tapered 'larvae funnel'. Play with the downward angle to adjust the flow rate - put a little 'slide' at the end to guide them to the beam. Otherwise, same thing but use a big piezo ultrasonic transducer as a 'hit detector'. I have some of those with a 'comparator w/ hysteresis' circuit (old 'knock detector' project) I could spare. Feel free to ask.
Are you sure? yes | no
Fantastic! For me, the main problem with easily separating and counting larvae is because "the larvae move, twist, widen or slip through any slit", and in my head I think it will be difficult to coordinate a hopper with a conveyor belt, for example (but it's just a perception for now).
Once separated, the "photo-interrupter" idea seems totally viable, regardless of the size of the larvae. Can you tell me more about the piezo ultrasonic transducer? Looks a fantastic idea as well. I will manage with prawns next month, and should be a good solution for insects and crustaceans :)
What about your work with bees?
Thanks!
Alex.
Are you sure? yes | no
I'm wondering if your larvae are mixed in with 'food-n-mulch bits' that will need to be dealt with. The 'beam-break' method might not work well - will count 'stuff' as well as larvae? The piezo discs (I have 1" dia. ones) generate a large-but-weak signal when they are tapped. They need a little 'FET buffer' circuit so you can have some signal left after going through the desired R-C noise filter - which then goes to logic input on your digital counter.
That might be the better method - you can have the larvae fall a short distance - to hit the 'tilted 45 deg.' piezo disc. - where they then bounce off into the scale bucket? Little bits of junk would not hit hard enough to trigger the counter?
On the imaging side... OpenCV is 'not real easy' - can suck weeks of your life away getting it to work like you think it should. Then there's that 'stuff that's not larvae' issue.
I think... conveyor belt - drop though 'junk removing breeze' onto vibrating hopper - to a steep funnel/slide (create a gap) - to the beam break counter ... would be my 'Plan A'.
My bee thing was just a backyard bee nest curiosity satiating project - wanted to subtract in-bound from out-bound to find out how many 'gatherers' were out-n-about. I fastened a little vibrator motor right to the interrupter thing - activated it with short pulse when bee broke the beam - it scooted them out - made a gap before the next one. :-) (Yes - the nest actually had an 'in door' and an 'out door' - made it easy. :-) ) Oh yeah... conveyor belts (rubber ones) don't have to be flat - you can raise the edges to make it like a shallow trough - keep larvae in the middle.
Are you sure? yes | no
Wow, Marlyn, very interesting. I will use diptera maggots without dust or food (I am using a sieve machine before).
Thank you very much for all your suggestions and ideas. I will begin to explore the "piezo discs" method, which seems the most reasonable. Let's play!
And very imaginative the approximation you had to the counting of the bees. I have seen some projects about it, but not as you explain it.
Kind regards all!
Alex
Are you sure? yes | no
To solve the wriggling problem count in batches. Take a snapshot and count. Dump. Repeat. Don't know how reliably you can count objects that overlap, but at least you have taken motion out of consideration.
Are you sure? yes | no
I think something like a 10x10 array of laser line sensors would let you see individual objects drop quite reliably.
Are you sure? yes | no
Thank you! Is it possible to use this system to count insects moving on a conveyor belt, even if they are very close together? Any suggestion?
Are you sure? yes | no
You would have to have the insects drop off of one conveyor belt onto another. Any sort of linear sensor array like a linear CCD and a laser would give a definitive on/off signal that could easily be processed to count insects. Too many insects becomes an issue, but probably not too bad. What is the acceptable error margin and how consistent in size are the larvae?
Are you sure? yes | no
The objective of this device is to be able to carry out quality controls, and to monitor some parameters during the performance of tests; in the latter case it would be the individual weight of the larvae (and their standard deviation). In the case of visual recognition, it could also include size as a control parameter. The size usually does not vary beyond a 15-20% deviation.
I suppose what you're suggesting is to install the sensor as if it were a beam of light, which will pick up the moment each larva passes through it. Or would it be a different configuration?
Thanks!
Alex.
Are you sure? yes | no
I think I can help you do this with detectNet, but you'll need a a decent budget for the equipment required. PM me if you like.
Are you sure? yes | no