Hacker News Reader

Comment by Havoc | Hacker News Reader

I don't see it detecting any actual weeds? The detection is just targets and the demonstration of punching the weeds is stationary. Which leaves out the key challenge - distinguishing between weed & crop.

View on HN

I agree, but to be fair, you could even skip that and just have a remote pilot looking at the rolling video feed and clicking on targets. It still beats having people actually walking under the sun and ruining their backs to pull out weed, or blanketing with herbicide.

I'm not so sure punching is a great solution though. Weed is incredibly good at reforming from very very small bits left standing.

beats ... I hope that was intentional.

Seriously though, you're right, even if it requires a human in the loop it's still a huge benefit. Farms could even use remote workers to do the identification.

This seems like the easiest way to get it working now. In fact, you could do the inverse at some point in the crop cycle - get workers/robots to identify the viable crops, and from that point on stamp everything else remotely plant-like out.

Humans are too slow. If this can be done at all, it can be done far faster than a human can do it.

> you could even skip that and just have a remote pilot looking at the rolling video feed and clicking on targets.

Then you could use that video to train a network to find weeds and get rid of the operator. You could also test the network a part of the same dataset to see how well it finds the right spots.

This will create weed breeds that look more and more indistinguishable from the crops.

Funny enough, this has may have already happened - Darnel was probably domesticated around the same time as wheat, and now it looks pretty much like wheat. There's a chance that humans accidentally swept it up when they domesticated wheat. The problem is that Darnel is psychotoxic and ingestion in the worst case can lead to a coma.

Possibly locked review: http://link.springer.com/chapter/10.1007%2F978-3-642-13145-5...

To get "more and more" you have to have a weed that triggers a partial response from the machine.

This works with a chemical - i.e. an accidental low dose, but I don't see how that would happen with this machine.

If, you had a weed that was so close to the crop it was only sometimes recognized then yes, but weeds mostly look nothing at all like the crop.

Kill them with fire! http://www.thermalweedcontrol.com/index.html

The problem of automated plant identification has been examined [0, 1] with good results. In [0] the classification is performed on footage from an uncontrolled agricultural setting, and the authors achieved a correct classification rate of 65% on tomato plants. Given that modern industrial agriculture is monoculture you only need to positively detect one variety of plant in an area. To avoid crop destruction due to incorrect categorization the robot could be used as a first pass and restricted to punching plants that are classified with a high degree of confidence. This could still reduce the amount of human labor necessary to tend to the field, but it might not reduce it by a degree that makes the robot a sensible economic solution. It may fail to reduce the labor needed, but I do not think that is likely.

I wonder how the performance of plant recognition degrades over the course of the crop cycle, if at all. We see in the video that some leaf matter remains after the weed is driven into the ground, and this leftover matter will influence classification during the robot's next pass. Unless the leaf matter is removed between passes it will accumulate throughout the crop cycle.

In [0] the researchers note that one reason they chose to deal with seedlings is that there is relatively little plant leaf occlusion at that stage. In [1] an end user is relied on to take a photo with a light, untextured, background and does not at all deal with partial plant leaf occlusion. That paper is not applicable to the uncontrolled field scenario.

I also found [2] while writing this comment, but have yet to read it. It boasts even better results (>90% correct for corn, 73.1% correct for tomatoes in an uncontrolled setting)than [0]

[0] http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.42....

[1] http://neerajkumar.org/papers/nk_eccv2012_leafsnap.pdf

[2] http://www.mdpi.com/1424-8220/11/6/6270/pdf

I wonder if there is an opportunity with GMOs to aid automated plant identification. For instance, insert a green fluorescent protein gene in your crop, and design a robot that senses the fluorescence.

Imagining the consumer reaction to green-glow-in-the-dark crops made me smile.

Multispectral imaging can identify weeds and separate different types of plants.[1] Humans have only 3 color sensors, but there's no reason you can't have far more. Some birds have 21. It's easy and cheap to do, although you need a camera with a special per-pixel filter instead of the usual RGB filter.

[1] http://www.bioone.org/doi/abs/10.1614/WT-07-104.1?journalCod...

I imagine if the robot is the one who planted the original crops it would know exactly where they should be. Any other growth is defined as a weed.

Exactly. You don't even need the robot to plant it; crops are planted in rows. Tractors and combines today already track exactly where the plants are planted down to sub-inch accuracy (http://farmindustrynews.com/accurate-inch -- other source, my family farm).

Reply on news.ycombinator.com