I'm life long friends with with family that owns a 80,000 acre grain and soybean farm, and they looked into drones last year or so for crop assessment, but found it wasn't really effective from either an analysis or price perspective. The drones didn't really have the range or endurance to properly sample the fields (we're talking fields of several square miles each), and the resolution of the camera just couldn't give you the detail you needed. It was basically just a waste.
Perhaps if there were more specific imaging sensors on the drones or something, you'd have something, but a simple visible light camera just doesn't reveal enough info.
I would like to have a conversation with you or them to convince them otherwise. I have been operating my own ag robotics company for 2 years and have contracts with biggest ag producers in Canada. 80,000 sounded big but not as big as biggest producers of wheat or corn!
Mavrx uses low flying aircraft to do the same analysis. It's surpringly cheaper to pay local pilots by the hour then to manage drones at this point although there's surely a point in the future when this changes over...
I'm doing work in venture capital with a specific focus on computer vision. I've looked into the agricultural robotics topic with some scrutiny. Here are some comments:
* feasibility of this new technology is due to the commoditization of three necessary components over the last few years: cameras (improved immensely in both quality and cost due to the smartphone revolution), drones/UAVs/small satellites themselves, and services like AWS that allow small enterprises to crunch huge amounts of geospatial data at little cost.
* there's good reason to believe that the commoditization of those components (i.e. availability and quality) will increase in the next few years.
* drones of the popular quadcopter type are a mixed bag -- their range is very limited because the batteries required to operate them are very heavy. This is tough because they have to survey huge areas. For large areas, many drones are required, which poses a significant overhead cost. However, what's nice about them is that they can be used with surgical precision: if something looks weird, drones can fly very close to the ground and take high-res pictures.
* conversely, what's nice about UAVs and small sats is that they can capture geospatial data on huge areas much more easily (though in lower resolution).
I think we're generally going to see a kind of two-tier approach, in which a high-altitude machine (a UAV mini-plane, or even higher up, a small satellite) does the broad aerial surveillance, and a low-altitude machine, i.e. a quadcopter drone, is sent out for random sampling to cross-reference the geospatial data, and for surgical investigation of abnormalities. The economic benefits of such a scheme could be quite significant for farmers.
I'm also looking forward to drones with more sophisticated sensors -- for example, heat or IR cameras. These are pretty rare on the market so far. Regrettably, they aren't being commoditized in the way that regular cameras are, so progress on this front may be a little slower.
"can capture geospatial data on huge areas much more easily"
What are you referring to as a "huge" area? All the work I'm seeing done with drones has been limited to ~700ha/day, if you've got perfect conditions. Normally it's much less.
What do you think drones with heat or IR cameras are most applicable too? When we've got machinery mounted sensors such as a biomass/NDVI camera or a weed-seeker, and the machinery has to make a pass over the field anyways, why not have those sensors onboard the machine? We're seeing Weed-seekers pay for themselves in the first year of use.
* AWS is way too expensive, and in most places where aerial mapping happens, you can't upload 30 Gigabyte to the internet faster than you can process it locally on a gaming PC.
* Smartphone cameras are almost sufficient, but mirrorless interchangeable lens cameras save the day.
IR camera modules that can be plugged into smartphones have been dropping in cost fast and the quality's superb for uncooled IR sensors, so they're not far off being usable.
I have to point out that this article is not factually accurate. TerrAvion (YC W14), is actually crushing the whole electric drone market and switching people who hold 333 exemptions and own drones to using manned planes for collection tens of thousands of acres at a time.
True innovation happens in amazing places. I'm really excited to see the future of agriculture enhanced by robots and AI.
One Bay-area company doing awesome stuff, not mentioned in the article, is Blue River Technologies. They make lettuce crops (of all things, lettuce!) insanely efficient with AI and hyper-fast distributed systems computations. This stuff is so cool.
[+] [-] jonathankoren|10 years ago|reply
Perhaps if there were more specific imaging sensors on the drones or something, you'd have something, but a simple visible light camera just doesn't reveal enough info.
[+] [-] pixl97|10 years ago|reply
One would think that IR would be very useful in diagnosing plant types and crop coverage.
>80,000 acre grain and soybean farm
I misread that humorously as a '80,000 acre soylent green farm' somehow. Guess it's bedtime for me.
[+] [-] Amir6|10 years ago|reply
[+] [-] robk|10 years ago|reply
[+] [-] hengheng|10 years ago|reply
[+] [-] johnloeber|10 years ago|reply
* feasibility of this new technology is due to the commoditization of three necessary components over the last few years: cameras (improved immensely in both quality and cost due to the smartphone revolution), drones/UAVs/small satellites themselves, and services like AWS that allow small enterprises to crunch huge amounts of geospatial data at little cost.
* there's good reason to believe that the commoditization of those components (i.e. availability and quality) will increase in the next few years.
* drones of the popular quadcopter type are a mixed bag -- their range is very limited because the batteries required to operate them are very heavy. This is tough because they have to survey huge areas. For large areas, many drones are required, which poses a significant overhead cost. However, what's nice about them is that they can be used with surgical precision: if something looks weird, drones can fly very close to the ground and take high-res pictures.
* conversely, what's nice about UAVs and small sats is that they can capture geospatial data on huge areas much more easily (though in lower resolution).
I think we're generally going to see a kind of two-tier approach, in which a high-altitude machine (a UAV mini-plane, or even higher up, a small satellite) does the broad aerial surveillance, and a low-altitude machine, i.e. a quadcopter drone, is sent out for random sampling to cross-reference the geospatial data, and for surgical investigation of abnormalities. The economic benefits of such a scheme could be quite significant for farmers.
I'm also looking forward to drones with more sophisticated sensors -- for example, heat or IR cameras. These are pretty rare on the market so far. Regrettably, they aren't being commoditized in the way that regular cameras are, so progress on this front may be a little slower.
[+] [-] fanquake|10 years ago|reply
What are you referring to as a "huge" area? All the work I'm seeing done with drones has been limited to ~700ha/day, if you've got perfect conditions. Normally it's much less.
What do you think drones with heat or IR cameras are most applicable too? When we've got machinery mounted sensors such as a biomass/NDVI camera or a weed-seeker, and the machinery has to make a pass over the field anyways, why not have those sensors onboard the machine? We're seeing Weed-seekers pay for themselves in the first year of use.
[+] [-] hengheng|10 years ago|reply
* Smartphone cameras are almost sufficient, but mirrorless interchangeable lens cameras save the day.
Completely agree on the rest btw.
[+] [-] jbms|10 years ago|reply
[+] [-] SixSigma|10 years ago|reply
[+] [-] CurtMonash|10 years ago|reply
Who knew?
:)
[+] [-] TheArcane|10 years ago|reply
[+] [-] eslaught|10 years ago|reply
http://ucanr.edu/sites/dcslaughter/UCDavis_High-throughput_P...
(Disclaimer: relative of PI.)
[+] [-] tuberry|10 years ago|reply
[+] [-] terravion|10 years ago|reply
[+] [-] terravion|10 years ago|reply
[+] [-] martian|10 years ago|reply
One Bay-area company doing awesome stuff, not mentioned in the article, is Blue River Technologies. They make lettuce crops (of all things, lettuce!) insanely efficient with AI and hyper-fast distributed systems computations. This stuff is so cool.
[+] [-] zjm5066|10 years ago|reply