Originally Posted By: kallog
Hi ImagingGeek.

I also work designing processor intensive software, so I'm aware of the limitations of computers. And I've done weeding on farms - distinguishing Italian ryegrass from perennial ryegrass for example. That's why this whole robot idea excites me. Sure it won't happen straight away, but oneday I hope nobody will have to do such sub-human work.

The whole farming thing gave me plenty of motivation to stay in school. There is a lot of boring, hard and repetitive jobs on a farm, and I'm happy to be away from that life (minus the farm gals, who are much more "earthly" than their city kin...).

I too would like to see a lot of this on farms, if anything, to increase the margins of my family who still work farms. It is not a profitable business, and its frustrating to see them work as hard as they do, for as little as they get.

Originally Posted By: kallog
Oh that's all you're talking about? That's nothing compared to the cost of building the mechanical parts without mass production.

But that cost does come down with mass production; the kinds of computing power needed though is currently expensive and will likely remain that way for years to come.

As I said, we do the same kinds of morphological processing here at work that would be needed to ID weeds verses crops. We make a point of regularity upgrading our hardware (its our single greatest material expense, after service contracts), but even with the most modern 8-core processors we're still needing to network several together to get the kind of processing speeds we need to do this stuff in a practical time-frame (and one not even close to real-time).

One day that kind of power will probably fit into my watch, but not anytime soon.

Originally Posted By: kallog
A GPS receiver is effectively free. I have a feeling you're trying to make the software problem sound expensive just because it's complicated.

Quite the opposite - the software is easy, as morphological analysis is pretty old-school stuff these days. Like I said in my last post, give me a bit of time and I could probably write it. The issue is the kinds of computing power it'll take to implement that type of analysis, at a speed sufficiently fast to be practical, and at a cost which is competitive with existing options. If the robot can only process a few plants an hour, it'll be useless.

Originally Posted By: kallog
In weeding I think there are 2 separate applications which you're confusing. One is to replace selective weedkillers that you spray over everything. The other is to replace manual weeding. I don't know what this is like elsewhere but when I was doing it, I was usually picking out a single unwanted variety from a field of one or two OK ones.

I'm thinking you may be a bit older than I wink Between GM crops and seed banks, farmers (at least up here) don't do a lot of selection on their crops. Even as a kid, which was a fair time ago (pre-GM, anyways), most farmers were switching to buying seed every year in order to avoid having to pay for the later kind of weeding you mention.

Originally Posted By: kallog
Choosing between 2 or 3 possibilities is surely much cheaper/faster than comparing to a database of thousands.

Not really. Its the actual morphological analysis which is computing-intensive. Once you calculate the morphological parameters, IDing the object takes milliseconds using fairly standardized cluster analysis routines (K-mean, or DBSCAN are the norm).

This is maybe beyond the range of this discussion, but in my own work I am trying to develop methods to automate the analysis of drugs and genetic mutations on certain cell functions that can be tracked with fluorescent tracers. The process is basically a 3-step one, separate the cell from the background, analyze the morphology of the cell and fluorescent tracers, and finally compare that morphological information to a huge database (currently ~50,000 cells strong, containing information on the effects of ~500 drugs/mutations).

The first and last step of this process takes less than a second each. The middle part takes anywhere from 10 to 15 minutes per cell. You can imagine, if a robot needed 10-15 seconds per plant to tell if it were a weed or not, the weeds would probably grow back faster than the robot would be able to remove them.