Counting potatoes

1

I want to count potatoes going across a conveyer belt in a mobile harvester.

I am wanting to count the number of potatoes as the harvester goes across the paddock.

So I am wanting to combine vision of the number of potatoes on the belt with a GPS position and then send the result over the 4G network to the owner’s mobile device.

I am thinking of using the AI capacity of Rasberry Pi 5 and the Waveshare coms card.

Any thoughts ?

2

Hi Ed
Welcome

Well, good luck with that. The last time I saw a potato harvester the conveyor was quite wide and I think from memory it shakes a fair bit to release any soil clinging to the spud. Don’t really know how you would go about “counting” them.

If you just want vision to send somewhere that is doable I think but actually “counting” as in counting the actual number I think the jury would be out on that one. You might be able to estimate somehow using the vision but the sizes would vary surely, which would make estimating the number questionable.

If you could get them to go down the conveyor in single file some sort of beam break or similar system would count them but would probably slow up the digging process to an unacceptable speed.
Cheers Bob

3

How accurate do you need your count to be?
i.e. are we ok with +/-5% or were are we needing +/-0.05%?

4

Hey @Ed290729, welcome to the forums!

Some great input from @Robert93820 and @Pixmusix regarding accuracy and feasibility, counting a few potatoes with low accuracy may be simple, but I would think that isn’t very useful.

That being said, you would be amazed by some of the capabilities of the new AI products from Raspberry Pi when paired with vision algorithms like YOLO and AI modules like Hailo.

See the below videos for some examples of people counting:

(These are two random guides I’ve found that somewhat visualise real-time object counting.)

Looking specifically at the 2nd video’s crosswalk demonstration. It performs relatively well considering the number of people that are largely obfuscated from the camera at any given time. I imagine, given a clear view of all potatoes on the line (such as a birds-eye view), a model trained on your potatoes would produce a result with high relative accuracy.

Curious to see where this goes!

5

I’m thinking +/- 1% the completed 27t load of potatoes gets paid to the nearest 10 kg.

6

You know what depending on your lighting and how much image porcessing youre willing to do i bet your in with a chance. Yolo pluggins, the good ones particuarly, provide a chart with error bars.

Pi 5s are like… 2010 laptop powerful. Amazing.

7

Thanks, I’m looking through the Cytron Technologies videos which then needs their guides to explain the details.

8

Hi Ed
Just a bit curious.
Why are you counting the NUMBERS of potatoes when you get paid by the kilo (weight).
Surely the number of kilos is the most important. Which you are not going to get bu counting the NUMBERS of potatoes.
Cheers Bob

9

Man looking at my old laptop from 2017 a new Pi5 16GB that is overclocked to ~3GHz would give it a run for its money.

1 Like
10

You can overclock a pi5 to 3ghz!?! :shaking_face:
Absolutely wild.
:firecracker:

11

3GHz is about the max, and will really depend on the “silicon lottery”, personally my Pi5 isn’t a fan of anything higher than 2.9GHz and refuses to boot with anything higher, but I have read tales of people getting up as high as 3.3GHz

12

Hi Dan
A classic example of stressing things to and beyond limits. All for what?? Mostly greatly reduced reliability as you have pointed out with the failure to boot.
Cheers Bob
PS anyway this is not counting potatoes.