I also have some insights based on a user study. Again it depends of what you want to detect, especially if its green on brown or green on green. For green on brown I recommend multispectral with NDVI / OSAVI , for Green on Green you need high resolution RGB with the MagicTool (soon to be improved in the comming weeks). Here is a more detailed overview:
Effectiveness of RGB vs. Multispectral Data:
RGB data can detect weeds as small as 6 cm in diameter but only at a ground sampling distance (GSD) of 0.34 cm/pixel.
Multispectral data performs better, detecting 6 cm weeds at GSDs of 0.58 cm/pixel and 1.63 cm/pixel.
Weed Detection Performance:
At higher altitudes (lower resolution GSDs), weed detection becomes unreliable for both RGB and multispectral images.
At 3.26 cm/pixel GSD (RGB) and 5.56 cm/pixel GSD (multispectral), weeds become unrecognizable.
Vegetation Indices for Weed Identification:
NDVI (from multispectral data): Highly effective at detecting green weeds in brown fields at 0.58 cm/pixel and 1.63 cm/pixel GSDs.
VARI (from RGB data): Some weed detection possible at 0.34 cm/pixel, but too much noise at 0.96 cm/pixel.
TGI (RGB-based): Too noisy for effective detection.
OSAVI (multispectral-based): Highly effective at 0.58 cm/pixel, but misses smaller weeds at 1.63 cm/pixel.
Conclusion:
Multispectral imaging is superior to RGB for detecting green weeds in brown fields, especially using NDVI and OSAVI indices.
Weed detection is highly dependent on image resolution (GSD), with high-resolution imagery (lower GSD values) being crucial for accurate identification.
RGB-based methods (VARI, TGI) are less reliable due to noise, making multispectral data preferable for precise weed identification.
Running an m3m in WA and finding 75m somewhere near the sweet spot for green on brown summer weeds, curious what you are doing with your overlap settings and if you have played with speed, and if you have seen any effects on accuracy.
I’ve just reduced the overlaps by 10 on each. So 60/70 instead of 70/80 I think it is. Speed I just set to max. It’s around 100ha/hour at 70m, less down time. A contractor told me I could do this and it’s been fine
awesome cheers, that’s a bloody good jump in productivity
Also noticed on thread above you are running a JD 11 section boom, same as i am, anything you have worked out for better or not? I have turned off boom pressure relief trying to get it to fire at better pressures on lone small weeds, and currently running pressure off at 400kpa/ 100L water, and finding max speed for sections to keep up about 19kmh, a bit slower for super high density areas
GSDs of either RGB and MS required to capture quality data of 0,34 cm/px and 0.58 cm/px or even 1,63 cm/px respectively make the job ineffective for M3M. Small scale jobs would be OK but would it allow for multihectar projects.
According to a table provided by yourself on M3M performance (Flighttimes & Speeds) both GSDs would mean flighing at max. 10 m AGL.
Am I understanding this right or am completely lost?
Do you use MS at 70 m AGL or RGB?
What drone do you use? Is it M3M? You started this thread asking about M3M and if so I do not quite undersatnd how you yield 70-100 ha/h when I read Julius’ table.
@willandratrust I’m have my pressure relief off also. You can increase the spray reaction time somewhere, 19kph is good, I only do 16kph. Also had some luck I thought if I switched from standard pump flow to low (tap outside near fill up). I’m wanting to demo a gen 5 screen to see if I can have smaller cells. I’m around 30% spray most the time so not the end of the world really.
Biggest thing I find is your cell size in pix4d, I’m on 1.5m now plus the buffer zone. At 1m the sprayer would occasionally not fire a section on single rouge weeds. Mine works really well if there’s always something going and firing all over the boom. Reality is they don’t like on/off variable rate I think
@pawel the info Julius has there seems reasonable. Reality is a 6cm weed is a seedling and with some good timing strategies you can wait until they are bigger and easier to detect.
I use a M3M drone with MS to create NDVI, i have active terrain follow on at 70m in most cases. I’ve been down to 50m a few times if I have time, any lower starts to get very inefficient on our scale.
The images can get abit blurry at this height but NDVI does a good job at finding them. I usually make it so I’m spraying some false positives to give me confidence I’m getting the smallest weeds out there. I don’t get 100% but it’s close (high 90’s).
many thanks for explanation and images.
Looks like this resolution is good for stubble - post harvest and detection of patches rather than individual weed plants. Which is v. good especially if you get close to 90% kill.
Will it work for post emergence in maize, sugar beets etc.?
Not sure on them crops. Only do wheat, barley and lentils here and if talking post emergence your green on green targets. The M3M isn’t good enough at these heights for this in my experience, it really is just for summer fallow green on brown. From my research at these heights you would need to go into a better drone and camera, like a matrice. The cost adds up quickly but you can put a 60mp camera on it and get under 1cm gsd at 100m height I think.
Yeah, thanks for your expertise and experience.
Indeed have been looking and researching how 61 MP camera could be useful for green-on-brown and green-on-green weed detection as well as some other jobs like plant count and phenotyping. I have been doing some flying for R&Ds with 20 MP camera and MS for a few years.
M3M peforms quite well but it has its limits and you pefectly picked up them and described.
These cookies are necessary for the website to function and cannot be switched off in our systems.
They are usually only set in response to actions made by you which amount to a request for services, such as setting your privacy preferences,
logging in, or filling in forms. These cookies do not store any personally identifiable information.
These cookies allow us to count visits and traffic sources so we can measure and improve the performance of our site.
They help us to know which pages are the most and least popular and see how visitors move around the site.
All information these cookies collect is aggregated and therefore anonymous.
If you do not allow these cookies we will not know when you have visited our site, and will not be able to monitor its performance.
These cookies may be set through our site by our advertising partner (Google).
They may be used by Google to build a profile of your interests and show you relevant adverts on other sites.
They do not directly store personal information but are based on uniquely identifying your browser and internet device.
If you do not allow these cookies, you will experience less targeted advertising.
