While we are thinking, they are doing it 
4 Likes
The laser is energy consuming and expensive. A point natural non-selective herbicide could be used, e.g. concentrated 
slurry. Works better than 
randup.
1 Like
Computers use to be extremely energy consuming and expensive too! Now look at them.
The laser machine looks very thorough, it does not leave any stragglers. If you could get every weed every time you might not need much energy once you get past the buildup of weed seed in an area.
But I think the laser is more suited to low vegetable crops.
2 Likes
You are absolutely right, to protect the soil and the creatures living there, not to pollute the groundwater with pesticides, to prevent pesticides from flying around, to use adequate doses of pesticides, and most importantly to leave a clean world to future generations. Now, big companies need to quickly develop their machines, see and spray, that is, spot spray applications, and sell them at prices that everyone can reach. If we are currently using 300 liters of pesticide (water and pesticide mixture) for 1 hectare, this value should be reduced to 50 liters on average with spot spray, so I think that 400 liters of warehouse volume will be sufficient instead of 4000 liters instead of huge spraying machines.
1 Like
Do you mean Roundup?
Yes, it was a bit of a joke. Until recently, I had a pig farm and when I poured fresh slurry onto the field the effect was similar to that of glyphosate. If it fell on anything green, it was dry after a few days! Hence the idea for a natural herbicide
Unfortunately, ASF (virus) and poor profitability forced me to stop pigs. In our company, now a kg of pork costs PLN 5 and the cost of feed per kg is also PLN 5, and where is the rest of the cost? Thick topping.
1 Like
I’m guessing slurry is high in nitrogen and acts in the same way that dog pee kills your lawn.
Just pondering that point… would urine work as an effective , eco friendly weedkiller?
1 Like
Wow! It looks like world is crushing the dreams of farmers everywhere. Where does all the magic free pork come from is a good question, and what does all the Ultra cheap pork eat?
Hey is there a dedicated group or thread for working on plant recognition and selective weeding? I’d like to start down this rabbit hole for a winter project, but want to avoid reinventing too much of the wheel. I found this labeled dataset of western Canadian weeds and crops. Seems like a good starting point for training a model.
Who is all working on this?
4 Likes
An open source selective weeding would be so good !!!
I currently play a bit with opencv but I don t have the level to go that far.
1 Like
openCV is generally used for object-object-vehicle plate-face recognition etc. It is successful in operations such as, but in weed operation, keywords such as tensorflow-cnn-rcnn-deep learning are the basis of this work, of course, a quality video card (nvidia etc.), a powerful computer, nvidia jetson javier card or google coral, and camera. necessary hardware, and phyton…
To read this, not so complicated, 10k of photos per target plants, which for me would represent about 15 plants, so 150K photos …
But the major difficulty in achieving precision seems to be the quality of the equipment especially so that it can see clearly at work, at 10-12kmh seems to be a challenge.
This study shows about 90% good at 1kmh, and only about 80% at 5kmh …
1 Like
What if we did it the other way around?
For example, if it’s a corn field, let’s teach the device only about the corn plant, and let the camera-sensor spray all the weeds it sees except the corn plant, maybe it’s ridiculous, just a thought… But the prices of spot-spray devices are still very high, 2m. boom is like 2000 euro, if you have a 24 meter sprayer it is 24,000 euro…
1 Like
I like this idea. Keep the good ones and clear everything else. It would probably be faster to calculate.
1K / meter is too high considering the savings made by targeting a farm like mine. Contact herbicides represent about 30 / ha, sprayer 32m (so 32000 for AI), 150ha, reduced the dose used by 70%, saving 21 / ha and it will take me about 10 years to pay for the commercial system.
For me, the right way to access this technology is open source and DIY.
The calculation is the same for the camera guidance of a row cultivator, another thing that I would like to find in DIY. This is why I am playing with opencv.
2 Likes
So is this kind of what you are working at?
I have a Class system now using stereo vision, but it narrow down the picture (compared to the above) The Class system only have 2 to 5 rows in picture.
But some day the Class system could stop working, and it would be better to go the opencv way
1 Like
It’s exactly that !
I started with a different approach from this one and I found myself in a dead end, and for a few days I have been moving towards this method … I had not found this article! Thanks for that, so I am taking the right path, although I am using c ++ instead of python, the method is the same, I will save some time.
Rather, I had in the idea a view of just a few rays, like your claas system.
1 Like
Just found this thread - some awesome site-specific weed control work here.
If anyone’s interested in an open-source weed detection device we recently (last year) launched the OpenWeedLocator (OWL) project. We’re trying to develop a community around the project, akin to AgOpenGPS so people have a base on which to build their weed detection projects and even integrate them with AgOpenGPS systems. @GreenOrpheus has been helping redesign the enclosure. Would love to hear suggestions/feedback from this community though on how to improve it or have people build a few!
3 Likes