Well, that good to see a good crowd there today. And it's great to have Tommy here because it was more of a process that it should have been to get you in here. But you're here. And so I'm sure you're getting introduced out in the countryside with that.
Excuse me, Is the fall cold going on here? But I didn't know how well you were introduced on campus here, so anyway, we're really pleased to have you here. A reminder that this is a cooperative effort between the our Institute for Digital AG and Advances items, Digital Forestry and Plant Sciences. And so we'll have a little time for questions. Make sure you leave a little time it all of it in the grad students are all saying, good luck with that.
Let's yeah, they know what you're talking about. Yes, this is right. Exactly right. So we'll say, okay. And before I forget it, I was looking through some stuff on my shelf recently and I didn't even recognize this book that was produced 24 years ago.
And this is called Precision Farming Profitability. It was developed at Purdue. It was a joint effort in the case of the New Holland Sigma side. And again, here is a section on precision weed control.
And so I know you got lots of new stuff for us today, Tommy, but some of the stuff has been around for quite a while too, so. yeah, definitely. I always like to say that agriculture has changed a lot in 100 years, but the one thing that hasn't is that the farmer has many problems, so everything's changed in 100 years and there'll be new stuff tomorrow too. That's right. Okay. Nice to get all excited in a way. Sounds good. Well, thanks, Bruce. Appreciate it.
So first and foremost today, thank you all for being here. I just want to let you know in advance. So my presentation today is kind of a hodgepodge of different things. But mainly what I wanted to do is highlight a handful of new technologies in the weed management world that we're either exploring on research side or exploring commercially or trying to more precisely manage weeds. So some of the things I would talk about today, we do have some ongoing research with some of the things are just really cool tools that I wanted to talk about that maybe in the future we can do some research with. And then I'm going to get into a little bit on the extension and teaching side as well, some opportunities there for digital AD and precision strategies as well.
So the first thing I did want to hit on though is just a little bit about me. So I am new to Purdue. I just started in April, so y'all may not know me super well yet, so I wanted to at least give you a little bit of background information. I'm originally from the bustling metropolis of Brownstown, Wisconsin, population. What we have through 262 when I looked it up the bustle All right, I would like to work on small dairy farms. Obviously, Wisconsin, right to get the milk cows every day.
And I do some small grain crops, things like that. All from there I went to the UW Platteville for my undergrad, completed that Bachelor of Science in 2012. From there I went to the University of Wisconsin-Madison to complete my Master of Science degree. As you can see, I was very well traveled through most of my life there up until that point. From there, I actually went out to Nebraska to do my Ph.D. and it was in western Nebraska that I was based on a research station.
So I got a little bit of the arid culture out there, dry land agriculture, which was pretty fun. And then from there I actually had a position down at the University of Arkansas where I was an assistant and then actually promoted to associate level as a faculty member there. I was in Arkansas about five and a half years or so. I got to work in soybeans and rice down there, which was a pretty fun crop to do something different in as well. So Arkansas agriculture quite a bit different, the Midwest. But then from there, just this year, I was actually able to join the ranks here at Purdue.
So I started in April, been here about six months. And when all the ins and outs of the Purdue Boilermakers nonstop. All right. I'd also be remiss if I did not mention the family. All right.
So I'll take show pictures of the cute kids that earns me some brownie points normally at the beginning of a presentation. So I just I've got my wife, Liberty, that's here in West Lafayette with me, as well as my two little kiddos right now, Brooker and Brinley, soon to have a third, actually, this fall. So fun stuff. But now just a little bit of background as far as why precision weed management is kind of taking the forefront across the industry. But we have some seriously problematic weeds in multiple different areas. You know, here in Indiana, I have a few just posted up here as picture.
So this is actually water hemp, supposed to be a soybean field. But if you can find soybean, I'll give you a dollar. Yeah, I'm not sure we can find soybeans in the water field. Giant ragweed is a big concern for us. A little bit of snow that we've mentioned there and cucumber, we've been getting a lot more calls on as well that can make harvest very troublesome weed management and things like that. And then I also like to highlight that these weeds are highly adaptable and can basically grow anywhere they want.
Right. So this is actually in downtown Des Moines, Iowa, where I took that picture and we had water hemp that was growing and thriving in the middle of a bustling town like like the morning. So we need some alternative strategies to help manage these things, particularly because herbicide resistance is becoming a big problem. Everybody loves this, this video that I like to show, but I don't have a better way to describe the battle that we're dealing with, with resistance other than this video. It's one of the best ways I can possibly describe it.
I apologize in advance. It is Arkansas football, right? I was down there for a few years, but I like to highlight this because this is what we're dealing with out there with resistance and weeds. So pig weed, as our example, we've got him charging up through there and we're throwing all of these different herbicides at him to try and stop it. And it just keeps on a chucking going into the end zone to reproduce, have more seed and we have to deal with it year in and year out. Right? We're just not stopping it. So we need some alternative strategies and precision.
We management is that is a big opportunity to help try and manage some of these weeds that just keep on moving. So that being said, kind of the first part of this I wanted to get into is just a few of the new tools that are maybe available out there that you might not have heard of on the Precision Re management front and then briefly discuss a few different research topics that either the weed science team here at Purdue has been doing or some of the things I've done in my past and then hopefully some future directions as well. So one of those things happens to be the sea and spray technology from John Deere that is commercially available right now. Thank you, Marcello, for driving the tractor in this.
It's pretty good. All right. What's the spray technology is is it allows in live real time for the system using vision guided camera guided systems to detect weeds as it's going through the field.
It's send signals right on the system to then automatically spray those weeds as it goes across a field so varied. It's live processing. Ben, so far pretty accurate, can detect weeds down to about a centimeter in size and the speed is really honestly pretty promising right now.
About that. John Deere is recommending we go anywhere between 12 and 15 miles an hour with the commercial sprayer speed, which is honestly really quite good for something like this where you have have to make as many decisions in such a short amount of time as this kind of system has to do. So this is just kind of an image of what it's doing. It's picking out all those different weeds. It sends a signal to the sprayer to say, I need to turn on and spray at those points. Right. So the reason this could be a benefit is, is for multiple standpoints.
But one, these kinds of targeted application systems have the potential for reductions in total herbicides used that, and that can help us, one, to maybe reduce any environmental contamination or satisfy some regulatory requirements moving forward and also hopefully can help our growers out. At the end of the day on their bottom line by reducing some herbicide usage and saving on input costs. So there's some opportunities there. Hopefully also it can increase weed control and whole farm efficiency from a few different standpoints.
Potentially. We maybe have less time required for spray or fill up since we're not broadcasting an entire field, we're not using the same amount of total spray solution. We can cover a few more acres excuse me. And then also maybe through less chemical flowing through the nozzles, we have a little bit longer lifespan of nozzles and other spray equipment, parts that our equipment can make it a little bit longer as well. So there's a lot of different reasons for us to try an approach to targeted spray, have it now just a few results from this system.
I wanted to highlight again, this is from Mr. Marcel Marcello Zimmer back there in the background conducting some of this research. And Dr.
Brian Young has had some funded projects through various organizations like the Indiana Soybean Alliance to conduct some of this research to find out if this tool one is feasible and also usable for a lot of our growers out there. So just some initial results that I wanted to help point out is on the left here you'll see the broadcast application. On the right, you'll see the targeted application. This left figure was herbicide settings and this right figure in spray solution, I should say, not in dollars in spray solution. And this right is soybean injury because hopefully we can reduce some soybean injury potential if we're applying chemicals that might be somewhat hazardous to the crop. And basically the main conclusion is, is that as you can see, if we do the targeted application for it versus a standard broadcast, we did see some spray volume savings and we also reduced our herbicide injury across the trial area by targeting rather than just broadcast the picture down here.
Sorry about the picture being a little blurry too, but you can generally see the the glyphosate control where we had Roundup Ready soybean completely uninjured by that application. If we spray something like Lactoferrin or COBRA, which can be an effective herbicide tool, but you can see it's pretty hard on soybeans, right? It burns them back quite a bit. If we can go to that targeted approach, those soybeans look a whole lot healthier and just have a lot more sportiness to them that hopefully we can have some opportunities there to reduce that crop injury, reduce concerns for consultants and growers. The other thing that this does is maybe opens the door for utilizing herbicides that normally aren't useful in our our different cropping system scenarios because we would effectively kill the entire crop, right. If we can make sure that we're us using very targeted approaches, we can open up the door and have some more options for herbicide rotations. And now I also wanted to mention John Deere sea spray is not the only vision guided sprayer on the market.
This is another one that we're working with here at Purdue, and Dr. Christian Kruse is up here upfront. He's also doing some work with them as well. So this is a robotic autonomous system that's right here in Indiana where it's vision guided once again, kind of like the sea and spray it finds the weeds, sends a signal, turns on to the spray, just where those weeds are at that.
Now, we did a little bit of research this year just looking at the potential for one, identifying those individual weeds species out there, the ability for it to turn on and then what kind of coverage it looked like around those individual weeds. So you can see that right here where we had a couple of different species. This is water hemp. This was a morning glory. We had Carr covered cards stacked around it just to see if we were hitting the entire target, what kind of coverage it looked like.
And then also based on the nozzles that they selected for this, what kind of droplet size is actually hitting that target, whether that's an effective droplet size or whether we needed to have some more discussions on setting up the sprayer in certain aspects to maximize effectiveness. The other thing that I love about this video that I wanted to highlight is zero easy to strip in the field right there. That's a that was on purpose. Okay? Not saying it was a good purpose, but it was on purpose.
Okay. What they did was, well, this year they wanted to leave a strip where they applied no pre emergent's residual herbicide and wanted to rely solely on the robot from this targeted approach to see if they could have more savings without having to use residuals upfront. What you'll notice is the circles back around the thing I want to highlight with these targeted applications or any of these precision weed management strategies, it does not get us away from using other strategies. The precision agricultural tools for weed management are just another tool, another cog that we can implement, but we still have to do a lot of our other things together with it. So whether that be using reemergence, residual herbicides like was used here and you can more selectively apply because you don't have near as many weeds up or using things like color crops to help suppress a lot of the weeds back and then use this kind of technology to go over the top later on or various other aspects.
But it's not an end all be all solution. We have to implement with other strategies to really truly be successful long term on the weed management side of things. But I was like, I like to highlight that with that video. It's pretty effective for that. So that's the sort of tech robotics side of things, the autonomous, a few other tools that we don't have or a current ongoing research with, but that are starting to make some waves in the weed science industry that I think could be some potential future options. One is the laser reader.
So this is currently commercially available mainly for vegetable and fruit productions, but it is again vision guided. So they have camera systems on there to detect the weeds. It sends a signal to emit a laser pulse to control weed. So that's what you can see here. That's the laser actually hitting a weed. That's after the laser destroyed the mast.
And so this could maybe be a great alternative option moving forward from this precision agriculture approach that gets away from herbicides to right. Again, having more tools in the toolbox, even with precision agriculture, is what's going to be important from a long term standpoint. Being able to use an alternative approach can be really beneficial in that aspect. And there has been some research on in the Northeast with this system so far and some of the reports I've been told is it actually is working really well. It does struggle with some bigger weeds and it's a fairly slow process.
So this machine only travels up to three miles an hour, right, versus hour, 12 to 15 with the John Deere Stansbury so limitations But but again we're laying the foundation to hopefully make it better in just a few years with some of these technologies. I will also say this is extremely expensive right now. I wired about trying to get one for research purposes at the university $1.5 million, and then it's also a $20,000 subscription fee every year for the software license. That's mandatory. You can't just like upgrade every five years or something because I asked questions and every year mandated subscription fee.
So right now, very expensive. But on the positive side, we talked about the spray, we talked about solar enough tech, we talked about the laser where there are a lot of startups, a lot of that, a lot of venture based companies right now that hopefully can start driving down some of these costs as we move forward, because there is a lot of interest in this area in agriculture. Now, not only do we have lasers and spraying systems, we also have some precision tillage pieces of equipment. Again, right now it's really heavily focused in a vegetable market. But again, these tools use some vision guided systems to detect crop rows and basically can tell anywhere and everywhere in between the crop as it's moving through through the rows. There are some road crop based precision tillage tools.
So a horse transformer tool is a really nifty camera guidance system that basically it's not looking for weeds, it's simply looking for crop. And as long as it IDs that crop grow, it moves its tillage times in and out to get as close as possible to that crop role. So we're not removing crop and costing us yield, but it can eliminate as much as possible in between. So we're making sure to reduce as many weeds as possible. But rolling through there and road cultivation standpoint, it also helps on the headlands when we reach those because it can actually pick up the times based on kind of an overlap sensor like we already do on precision planters.
But then it makes sure that we're killing the entire area that we need to, but we're not pulling out our crop as well and reducing yield because we've just eliminated some of our crop. So that's pretty cool. Me Tool. Also, there's a tool based out of Australia, maybe a little bit less effective for Indiana, but still kind of cool to talk about. It's called the Australian wheat chipper.
And basically what it is is again, camera guided, you can see what the field look like beforehand. They're dealing with a major problem of wild radish, which has a big, you know, taproot that they're trying to remove. And so what they do is when the system goes over it, it sends down this all, see it right there, the whole plow shank to basically just rip out that plant.
And it just keeps driving through the field. So this is the beforehand, this is the after and after photo where they didn't have to kill the entire area, but they could just chip out where all of those weeds were at possibly an option where maybe we need to try and maintain some more minimal tillage options or other various things that. Yeah, and not a broad scale solution, but possibly in some areas they would have a fit. So lots of potential to try.
I can program these so I could be, let's say young corn, soybeans, rice, whatever. I assume this is not a single plant or we haven't gotten that. Well, yeah, I think that the challenge that tillage is a good example so the tillage one like for this one here right. It's basically it's it's just looking for any kind of row. So it's not necessarily crop specific per se, but this one is old for crop. This one can handle if there's a crop in there, it's it's a it's a free cropping if we're talking say the John Deere see and spray that is very crop specific.
So they have specific models that you can use that in corn. There's a lot of corn, there's a model for soybean, there's models for cotton, which is very handy for Indiana. And then there's a model for fallow matching.
So you couldn't use it in any other crop technically, Right? So an attack right now, they are basically kind of corn and soybean. They're looking at it and some other crops as well. But mainly it's corn and soybean right now. So it depends on the technology, the laser wider.
I'm not exactly sure how that ones go on, but it depends on each each system. Each one is kind of unique. Yeah, it's a good question for I thank you.
Feel free to ask any questions as we go throughout too. I'd rather have a discussion. I'm good with that, so please feel free to ask questions as we go.
So a lot of questions I have in terms of accuracy, what form do you think would be the most accurate at the moment for controlling like between specific models for crop models? I'm pretty certain that you have tested. I don't know if you have some of those. So that's a that's a great question. So the question was what kind of the most accurate, right. As far as the different systems that we've had? I mean, I would honestly say from the ones that we that I've tested and talked with and stuff, if we're talking precision removal of weeds, the John Deere scene spray is probably, if not at the top, near the top, but they've been working on that system since, you know, 2016 or before.
And machine learning is all about beating images, right? The more images you feed through it, the better it can learn, the more a better understands that stuff. They have fed tens of millions of images through their their system at this point that they have trained those models really highly effectively. They've also kind of developed a hierarchical structure of how it makes decisions, which has become really effective at basically deciding whether something is a weed, whether something is a crop and go up or down.
So I would say if we wanted the absolute highest accuracy and precision right now, that one seems to be top of the line. Now, if you want a little bit broader scale, like I think this is a really nifty tool that can be used kind of across the board because it's not being as it's not as precise, right? It's more just let's find the crop grow and we'll eliminate everything else out in between. If that's kind of a goal, that is a really effective tool at that aspect. So it depends a little bit on your objective or goal, but that's kind of where I would say things shake out a little bit.
Yeah, good question. All right. From there, are there any questions on some of those new tools or research opportunities, things like that are? The next thing I wanted to move into a little bit was talking about imagery and drones and using some of that technology for either detecting herbicide injury or crop injury from various weed management related strategies, or also trying to detect weeds and set up some targeted approaches. Correct. So the first and foremost thing I've got on here is actually not even dealing with drones, but I figured I needed to highlight it is there are actually proximal sensing techniques that we can use in the weed management world. We did this with study just a couple of years ago where we used plants in the greenhouse and conducted a digital imagery analysis with them right off.
If you listened to Lee Miller give a talk a few weeks ago at all, he talked about using turf analyzer. This is very similar. It's field analyzer. It's kind of the same company, but it's based more for kind of a field track side of things. But basically this program, we're able to take pictures and as long as those pictures are based from a fixed height and it's standard across the board, the area that you're imaging can be the same and you eliminate all the hues and saturate motion picture of pixels that aren't what you're targeting. Right.
So this was the before picture that I uploaded in the system. This is after I eliminated everything in the background and that it gives you an estimate of what the canopy coverage is based on. Those pixels are kind of a rough size of that plant.
So here when I would factor anything out, the result was 100%. So actually that's factoring in everything. Once I started eliminating everything, it basically told me that the soybean plant was taking up about 30% of that imaging, roughly speaking. Now, the way you can use that, as you can start making some comparisons with non tree controls and growth patterns over time with various treatments. So that's exactly what we did over here.
So this is like canopy coverage reduction percentage. So basically we did a drip study out in the field and we had live plant samples to evaluate herbicide drift and and the impact on the plant. So each we had a planet in each one of these downwind stations. We took these pictures at four weeks after application and we basically just did a percent reduction in canopy coverage from our non tree to controls that had we were able to grow the entire time without any exposure to herbicide. And you can see we were able to fit a couple really excellent curves to that data on estimating the base clinic, that growth pattern and the effect of herbicide injury at those various stages.
So the black line was actually from an aerial application, the blue lines in the ground application. And so we were able to get solid, useful quantitative data rather than just visual observational data to use and estimate how that herbicide was affecting our crop. But now we also can do some of this from a remote sensing standpoint using drones and other aerial observational tools, satellite imagery, things like that. So this was a study we conducted down in Arkansas just in the past year or two, and we were looking at herbicide, a new herbicide that was being registered for use in rice. And we wanted to look at the crop injury potential from it.
So we had our field study set up once again, we're using field analyzer and it's a very it's a nifty program when it comes to a field research trial because basically you can draw on your grid of plots, remove the borders, if you have any borders and things like that. Again, you do the same system of eliminating any of the background just so you get down to the crop that you actually want to evaluate and it'll give you an estimate then of that canopy coverage or potential loss due to herbicide injury. So on the left again, hundred percent coverage. If we look at here, I've got plot one on one highlighted that is same and it's tough to read, I think it's about 54% canopy coverage estimate with that plot and it all at one time you can spit out all of the data for your entire trial for canopy coverage estimates and once again we are able to use that data over here in the publication to provide estimates of those various treatments and what level of canopy coverage we have over time. So you'll notice this treatment right here that basically wiped out the rice.
We had a drastic really delay in what that can be coverage look like and it did eventually catch up. But you can see again, quantitatively rather than just qualitatively, the numbers to looking at these various injury symptoms that were out there from the herbicide. Now, I like to highlight this is a little bit different than if you listen to Doctor John Jim last week talk about his precision sensors on the ground. Right. I think like what the discussion got to at the very end was a really useful discussion is there's room for both of those technologies in our in our weed science world right now. This is a really for me and we're going to get into a little bit more of the remote sensing side, but this is a great way to get a better spatial representation of what's happening in the weed science world, whether it's weeds or herbicide injury or other things, but it can't really tell us a whole lot of specifics about it.
Right. Gets us a rough idea. We can a consultant could go talk to their grower and say, Yeah, we're seeing some significant hotspot injury symptoms out here. Your canopy coverage might be reduced by that much, etc.
But if you really want a precise answer, then you need to go and do some individual plants. Those more high tech sensors that Dr. John Ginn was talking about, things like that. But this is kind of a nice starting point to lead in to give him more direction on the specific side of things. So we kind of go hand in hand together. I think for a lot of us. All right.
So though both of those things, we're with the Fuel Analyzer program, which is really a relatively base program. It's all run in Java. It's pretty simplified. There's also a program called Pix 4D, which we're running some tests with now to read into a whole lot more specifics and look at a lot of different index values to try and maybe get a little bit more precise than what we were just talking about. We'll never be as precise as some of those individual sensors, but if we can find some different index values that get us a better answer, again, the more accurate we can be with a system such as a drone like this, that really takes a minimal amount of time to fly. And we have a lot of growers and consultants that are interested in this kind of technology.
From a speed standpoint, we can find some ways to make it more usable. Everyone will be better off. So just a couple images from some studies this year. This is corn at about, I want to say the V5 or V six stage. And so you can see from an RGV image, there's not a whole lot really that we can detect when we start looking at all those crop.
And then if we start breaking it down into different sensors, though, this was actually using a thermal sensor on one of the pneumatic three thermal drones and we start eliminating out some of the background noise. You can see the crop rows a whole lot easier. And in the system you can actually draw out every single plot like you see here to then spin out values spatially for each plot on what that that is looking like as far as treatment of X.
But now you can drill down into other different different indexes. This was this index right here is called an M quarry index, which is basically a chlorophyl absorption type estimation. And this one has been really useful, I think, at least in my initial tests, where you can drill down and eliminate a lot of the different noise out there, like the weeds and things like that, and get down to just those crop rows. And again, each one of these plots is a different treatment and you can start to see where some of these have a little bit higher chlorophyl some of these have a little bit lower Chlorophyl There's those small minor differences within those cornrows started to pop a little bit more with some of these various indexes and if we actually put the numbers to it, show a graph. Sorry for you don't like looking at graphs, but I throw up here a little bit, at least for a year. Yeah, that's right. You can see here that what we looked at those images and put them in here.
You can start seeing some of these trends with the different treatments where on MDI side of things, we've got the ones that are higher and greener and more healthy versus the treatments where we started losing some of our MBI and they were potentially more crop injured. We can also see basically the inverse reaction on the thermal front, which makes a little bit of sense that stressed plants are going to be a little bit hotter than non stressed plants. And then here's that am Akari one which even kind of again not by a lot but thinned out some of the areas just a little bit. We're going to get just a little bit clearer picture but it matches very nicely with the and the high grade and this is actually Jada's project back there that that I was just piggybacking off of him. So we had some discussions about which treatments here were falling at the bottom of these.
They wind up with a lot of her visual injury ratings where there's about three certain herbicides or adjuvants mixed in that were causing issues. And this gave us a pattern, a trend again quantitatively to say, yeah, those those herbicides are those adjuvants are the ones causing the most injury with this specific herbicide that we tested. And I think someone had a question. So these simple things, for example, going on and work to the up when it can work effectively were just about anything.
So actually next slide. Soybeans. Yeah good question though so picks 40 is is really effective across the board. I actually hope to do some work with Steve Meyers and his floriculture weed science group to maybe chested in some of the different for crops as well for picking up various things but in effect what it is is it's that software is just as good as whatever imagery your collect. So if you collect thermal imaging it can handle that.
If you collect a multi-spectral imaging, which is where that Campari came from or came from, it handles that it can handle just our imagery, but each one of those will tell you a different thing. So the picks for software can handle basically just whatever imagery that you're going to feed into it, and it gives you answers based off of the resolution on that. Yeah, it's a good question, but excuse me.
So I didn't want to just give an example of any trial to here on the left is just the RGV image. Here on the right is where we calculated out the inquiry again. And just what I like to point out is if you look at this RGV image, you really can't see a whole lot that's going on, especially with soybeans that are about to be want or need to stage. Right. And also you'll notice here and this is one of the things that Dr.
Jain talked about last week, it can be complications with some remote sensing data, a lot of shading happening here, right where we had some overcast cloud cover that happened, which makes some things more complicated. If we start picking out some of these different index values in these softwares that we can use, we kind of effectively eliminated that shading problem. And you could start picking out the soybean roles a whole lot more and figure out which plots have higher chlorophyl contents have higher growth patterns with the soybeans. Even at this early stage around, do you want to be too good? So again, just some really cool opportunities here, I think, to pick out various things very quickly because with this song with the Magic three drone that I was using in this, I could fly these areas in about 15 minutes and cover it with one battery. Well, that's a big deal, especially if we start talking commercial consultants and agronomists. That's now on top of that, I talked a lot about basically idea in herbicide imagery there.
We can also do the counter aspect to this and not worry about the crop. We just completely eliminate the crop on the imagery and we start detecting the weeds specifically to hopefully do various things. So once again, this was the initial image, the RGV image, and I just want to match up with when I started playing with the settings on what I was looking for at the bottom, I effectively eliminated out all of the crop and I just got where our hot spot patches of weeds are now not 100% accurate by any means. I mean, like this is an event where you barely contain anything, where it's a pretty good patch and weeds right there. So I'm not trying to say this is an end all be all answer either.
But again, for some growers and consultants and things that are trying to get some rough answers out there or even from a research standpoint, if we're trying to figure out where some hotspots are for projects or just about, you know, what kind of biomass do you think we're dealing with, this can be a tool again to give us some quantitative numbers in a quick fashion. I also will say one of the things going back to the sea and spray systems that have been a big challenge for applicators using that, they often question, well, how much do I mix if I don't know how much I'm going to spray? Right? This might be an opportunity where they could fly it, get a rough estimate, at least that are probably going to have about 50% of the area covered in weeds. I'll mix for 50 to 50 5 to 60% of the field and it gives them a better judgment call on how much to be mixed, how much spray solution to be mixed in for certain sort of fields.
But then with that, the other cool thing I wanted to mention, where we're already at with some of this precision technology is basically creating our own do it yourself prescription spraying from some drone imagery technology. Right. So I'm just going to walk you through it because I think it's kind of cool to see it in action.
You know, I can't show the actual program. I got some screenshots to show you. So this is just an image that we took again in the trial this year up in Francisville. And we wanted to I wanted to just play with it and see if I could detect where the weeds to spit out my own prescription map. So the paint is the border that I set up. So this is the field that I'm concerned with.
Once you go through this magic tool again, it's in this picture for the program. It has an AI algorithm automatically built into it that you basically want to tell it the things that you want to keep and the things you don't want to keep in the image sample of that. So all the blue circles on here are the things I told it to keep, which is when I zoomed in to those pixels, those are the weeds, right? And then I started highlighting, okay, the yellows I want to get rid of.
I don't want to keep soybean plants. I don't want to keep bare ground, but I don't want to get rid of all of that. So I'm not spraying any of that.
And in this example, all I did was I selected 22 cells in that entire image, 22 cells, ten. I catch 12 I got rid of. And from there I was able to spit out this map right here where it literally picked out all say all but it picked out a bunch of the weeds. And again, a lot of the hotspots where they were at to create a prescription map of here's where you need to go target From there you can it allows you to add a buffer because we all want to have a little bit of, you know, error potential, right? So you add a buffer to it.
So I did that. And then even at that point, even after I added a buffer around all of those areas that it detected when, it spits out to the very end here. It'll tell you your total area within the field, but then the total area that is going to be applied and so on instance, when I did this little demonstration, I had about a 50% savings. It went from about a two and a half acre field down to about a 1.10 million acre area that I was going to spray. That's pretty neat.
And this this this whole process took me about 10 minutes to do. I'm all about user friendly when it comes to any of this stuff. Right? So about 10 minutes I was able to spit that whole process out, which is really, really nice. Now, the downside is and some of the things that we have to test on the back end is how effective is this prescription map when we start running it through various systems? And also the accuracy was variable, not going to lie ahead. So if you look at the zoomed in image here, you'll notice a whole lot of purple where it's going to be spraying and there's not a lot of weeds in that area there. Right.
And then also, if you come right over here, there's a ginormous one right there that I didn't pick up and it missed. So by all means, not 100% accurate. It needs to be fine tuned.
But the other thing with that guy tool is it has the ability to learn it. It gives you the option that you can check the box and it will learn from your selections and do a machine learning approach in the background. So if you did this enough, you could probably train it to pick out exactly what you want and be a whole lot more effective. That yeah, that's pharmacology, which is more than just machines, but right exactly What what we expect in control.
So I'm going to have to expect some excuse. Yes. So this is what I mean you're right. Well, I think that's where that puts me. I think that's with any technology at this point, it's any one technology by itself is not going to be 100% effective.
And so it's putting together all of the pieces that can get us there. But you got to put them together and not just rely on one thing, whether it's a piece of precision equipment, you know, machine learning, A.I., etc., or whether it's, you know, the old herbicide allergy, you know, cut.
All right. With that being said, sorry, let me back up once again. But that being said, this prescription map can be spit out. Numerous aspects, which is pretty cool. So you can spit it out to go to a spray drone and export it to you and say spray drone or DJI spray drone and a drone can go spray it, which you'll see we'll talk about a second.
Or you can also just spit it out into a shape file and that shape file can be plugged into pretty well just about any standard ground sprayer now at this point that has the pulse with modulation. So the solenoid valves at each nozzle that you can precisely apply, which is a lot of our sprayers now, you can spit this prescription map out and make your own do it yourself prescription map sprayer with a lot of the technology that's already currently available. So this makes something like this a whole lot more feasible again to growers rather than having to spend a half a million or $1,000,000 on a piece of equipment, you have about a $2,000, you know, subscription fee or so for kicks, 40 and everything else just runs on the equipment you already have. But with that being said, that runs into the whole spray drone conversation though. Mike I spent a lot of time on this, but there are a lot of question marks in this world, this technology in the weed science world right now, too, as well as in the plant pathology world.
Don't worry, nasty only be out. But so there's a lot of question marks on on how effective these can be, whether they're you know, they have big whether we can control weeds and other things like that. So just a little bit of research that's been done on this. And Hunter back there is our expert drone graduate student.
Right. Okay. So but so this is actually I stole that video from his project this year, but this is actually a project that we did down in Arkansas last year. And we were just looking at the weed control from the bottom drone at two and five ounce breaker, comparing it to a standard round rate of ten gallon breaker just to see if it had any merit at all, if it could stand up on its legs and be effective, I'll admit I was trying to make spray drones fail when I set this experiment up, so I gave it a contact herbicide Paraquat, which is notorious for needing higher spray volume, better coverage, things like that. I put it at a low rate and then at the low GPA and I'm like, We're going to make this fail and that'll be the end of the conversation.
Well, it actually somewhat worked a little bit. And so now we've got a deep dive into more research. I guess that's job security, more research opportunities for the future.
But the yellow boxes were the two GPA treatment that you can see where the blue boxes were, five GPA from a drone. Maybe a slight difference between those two, but honestly, not a whole lot. And then the black boxes are the ten GPA treatment.
And it's really tough to pick out, particularly between the five and ten. Any differences visually there? If we look at some of the numbers we looked at one week after we applied ratings, the ten in the five GPA treatment, so the drone at five versus the ground rate of ten. No difference in weed control. The two GPA lagged behind a little bit, but not not a terrible amount. But let's take behind a little bit at four week after visual ratings, again, same story. Five and ten GPA were identical.
The two GPA lagged behind a little bit, but not terribly much. But if we get out of visual ratings and we do some quantitative assessments, the weed biomass at the end of the four weeks was statistically similar. We had the same biomass across all treatments, the very which is another remote sensing index using an RJ B camera. So it's basically a measure of greenness of the plots is effectively what that was showed at one week and somewhat matched up with our vegetable ratings.
We had about 75% or so greenness reduction in greenness, all four, five and ten, and that two GPA lagged behind just a little bit. So that kind of matched up with our visual ratings. But then that greenness index matched up almost identical with our biomass data.
Four weeks after application. So what all of that tells me, right, is that sometimes our eyes can be a little bit deceiving when it comes to doing visual ratings, which makes sense, but also what it tells me is it looks like there may be something to two GPA with weed control, maybe some still complications here, but it's more feasible than I originally estimated with we're trying to make spray John's work from a control standpoint, even with a product like of which is a major contact that Now the complication we just talked about, though, is the factor of the swath width. Okay. And where this spray is actually landing on the ground and what's expected.
So these were some of those treatments. We did this work here at Purdue, but this matched up with that study in Arkansas. So we had an excellent LFA nozzle, which is a fine spray at two GPA of five GPA.
And then we had a wind farm A to the. Yeah. Which was a larger drop of size, like a very coarse, extremely coarse of two and five GPA. They may notice a problem with those swaths. So this is dead center.
The spray drone flew right down through these. We have 15 feet to the left and 15 feet to the right. And we were supposed to have a 20 foot swell.
That's what we told the system should be 20 feet wide. The problem is, is that every one of these systems did something different, basically, for what? That's what was right. So the zero two GPA, we basically almost have a full 30 foot swath when we slowed the drone down to get to five GPA, that's how we got to that range. We just slowed it down. It basically forced all of that spray into the center rather than spread it out. And so now we only have about was that roughly a 12 foot swap.
And it's also not centered. It got shifted slightly because of, you know, a little bit of a wind direction. But if we went to a larger droplet size, slowing down didn't matter as much with the GPA, but the larger droplet size is the same thing here and all funneled into the middle. And that also did not get us our full 20 foot swath.
We ended up with only about a 10 to 12 foot swath. That becomes problematic from multiple standpoints. If we're doing an entire field broadcast, we're going to streak it with this shift in the direct center of where things are getting happen.
If we're doing the targeted approach, that also becomes a problem because now where do you fly to target? Apply to those different weeds or those patches? Because you're probably going to miss some areas if that spray is shifting. So still a lot of research to be done on the spray drone front to some potential. Well, a lot of kinks to be worked out about keeping that same height of your line so that with this one. Another good question with this one, we kept the same height at all times.
It was always eight foot for our study. Some people fly 1012. It varies a little bit. Anywhere between 10 to 15 is common. Normally, if you keep the same height like we did here, you'll still see this effect.
If you start changing the height, this effect will change. But so at the end of the day, to find the right answer is it's going to be a combination of messing with all of those things. So get the correct swap. But no one has that answer right now on where you need to move your height too, and what the locks should be and everything you want.
Problem that we had at the beginning was that we thought since we have program, the drones to specific time, we started seeing that the things started to drop. We're beta and come back Yeah and so but there are ways we can fix that it make you want to excuse me some of the older spray drones we've noticed have that problem a lot where they drift a lot in height. Some of the newer ones seem to have better they have better radar capabilities. Now they've upgraded that and so they're holding their height a little bit better. So I had some conversations where like the new TI 50 from DJI have a much better radar sensor that they're able to keep that height a whole lot better. I'm bombing a ground station that helps a lot. Yes.
Yeah. So, you know, so I mean, it's all variable, right? Everything is a new variable comes in all things, so it's awesome. But with that being said, that's kind of the research tools.
Part of that I want to talk about today. Are there any questions on that front? All right. As a three way appointment, I'd be remiss if I did not talk about the teaching and extension side at least a little bit when it comes to some of these digital tools as well.
Right. So I want to highlight just a couple of things. It's not a lot, but a couple of things that we can be using some of these tools on that front as well. So one of the things I wanted to briefly talk about is this summer I did a study evaluating some of those pre public air chat bots and how well they were at identifying weeds species right? Like every county extension agent or extension specialist gets all these images and they have to try and ID those weeds from that image.
Can these air wasters right and do their job. So with the weed ID, I sorry, I should say I tested these for free ones available out there. So Chad GBG, Gemini, you and I, naturalists and I just use the free versions. I didn't pay for anything.
I just want to see if someone wanted to test. Don't get a result. So this table up here shows the results from a weed ID standpoint. I ran 104 different pictures of weeds through the system.
The one takeaway that I got was that I naturalist is actually pretty decent, right? So I broke it down into various categories On whether it was 100% correct. It nailed it. If it was completely wrong with what it told me, if the initial guest was wrong, but it threw out a couple alternatives and that was right, if there was no answer at all, I just gave up and said, I don't know or if it kind of got it down to a genus or family, but didn't narrow down to the exact species.
That's kind of how I broke it all down. I naturalis basically it got roughly the right answer. 82% of the time, whether it was the right genus or family or whether it had an alter in that. Correct. Or whether it was just correct right out of the gate, I naturalist about 80% of the time gave us a near right answer. If you compare that to some of these other ones that are publicly available, it starts dropping off significantly as far as correct answers go on chat, CBT and Google Gemini were fairly equal. You just likes to give up a lot and says I just don't know which.
Sometimes that may be the better answer right then given me the wrong complete wrong answer. So maybe not necessarily a bad thing, but a lot of these can't be trusted from from an extension standpoint. And so one of the things I like to say is we still all have jobs as plant botanists and biologists, right? And extension personnel. We still need each one of us to be able to do some of the stuff. But there's a lot of work that can be done on this front, too, with developing tools for the extension service to help better IP weeds When it comes to herbicide injury, I also wanted to evaluate that on whether these different systems could I.D herbicide injury. I'd argue US doesn't do that, so I had to drop that out.
But I tested Chad's utility, Google, Gemini and you and I only had 26 observations for this or images like that at this time. But what I was really pleasantly surprised with is Jackie Beatty almost knows what it's doing when it comes to herbicides. It was right about 73% of the time. And this one I had broken down into similar things here, except I would initially ask it what site of action is this? So I give it a broad based question. First, If it got the site of action right, I would follow up and be like, okay, well what specific herbicide caused this injury? And Judge Beatty, almost half time I was able to get the exact herbicide right, which I was pretty impressed with. And then the correct side of action was No 23% of the time.
So I honestly did fairly well in idea and herbicide injury. You once again was just like, I don't know, I give up a vast majority of the time. It didn't want to give me an answer. Again, maybe not always the worst case scenario though. Google Gemini was the massive disappointment from a weed science perspective.
It only gave me the right answer about 30% of the time. I'm pretty sure a plant pathologist may Google Gemini because it wanted to give me plant pathology answers all the time and I'd ask it what herbicide symptomology causes. And it would be like it's not a herbicide, it's this disease. I was like, Darcy. So, so Google Gemini really strong. So basically the conclusion from all this once again is we still need a person in the seat to handle a lot of these questions.
At this point, there's a lot of room for improvement on these things, but I also like to say that each one of these has its own kind of unique AI model, that it's running in the background and it'll give us some very stark different answers for some of these things that we feed into it. So being careful from an extension and education standpoint is pretty critical. There. One thing that I wanted to just kind of throw out there as an upcoming thing that we're doing is we're going to have an upcoming weed control technology survey to to release to the public and try and gather a lot of different insights when it comes to these different technological pieces from a weed control standpoint.
And I think some of these questions and answers will be really beneficial for a wide variety of us in the room, because we're going to also ask things like What is your Internet service that you currently have available right now? What does your cellular signal look like and what's the capabilities for you to implement some of these different technologies with our, you know, Internet and cellular providers? You know, what is your return on investment look like for some of these tools? How much do you what we want to spend on various things like that? So it's going to be a weed control focus, but I feel that it will have a lot of utility across the board for various other disciplines and extension personnel. So I just want to put a plug in for that. And we're releasing that this fall winter hour to try and gather some insights across the board there. And then just kind of finishing up. One of the last things I did want to mention is I didn't want to say a big thank you to the Purdue wheat science team, one for just welcoming me in right to the group and the team. I've enjoyed that a lot.
I will also say to you, this group, whether it's the row crops side of things with Dr. Brian Young and Bill Johnson and their crews on the Hawks side, Dr. Steve Myers, Dr. Aaron Payton on the turf grass side.
Everybody's reached out to me wanting to do various things, bring me into the fold and it's just been an awesome team to collaborate with and every one of them is looking for answers from an extension and teaching standpoint across the board, which I really appreciate for all of this. So it's been a really important aspect of all of us and it's it's been a cool thing to be a part of. And so then the final thing, I never walk away without use another football reference, okay? If you take away nothing else from my presentation, here's a digital tool that we can use on the extension education standpoint with trying to make our points land home and get across.
So my final message using some digital agriculture here is how we can get long term successful weed control with these different tools. And so we play the video. The first and foremost thing I like to mention is it's always going to start up front with the big boys, making sure we're starting clean and using full rate residual herbicides, right? We don't get away from that. We need to implement these other tools. We also need to do the little things in the background not to drop into coverage. Otherwise the quarterback can just drop it off and it's an easy completion.
So we got to do the little things right. But then where technology comes in, it can play a really important role as we fly in off the edge here with the novel technology macro, which we call the funnel, and we follow up with integrated we management, all of these things working together as a team to really provide long term successful WE management strategies. But I will tell you two, this was a tough find for Purdue football last year, a very positive aspect. Anyway, that's that's my final conclusions. I just want to say thank you for coming today and I'll take a hope that you ask questions for time.
I have several here. If you don't know how to keep me from asking, go ahead pray. Tell me on the mispronounce it so the so an attack. It took me three months before I got it so so look into it but how many gallons that thing go You know what would take five of you.
It's got two tanks on it like one on each side just to balance it out. I'm pretty sure total it's 40 and gets 20 and one. That's all. Yes.
And then we fill up so it stops at 13 to come fill that thing up, right? Absolutely not. So it doesn't go back somewhere to right now at this moment in time. Yeah. Somebody has to go out and fill it, but they are already working on that. It has an an automated return on it. And basically they're automating everything on that front.
Like if they showed me a video of like their initial prototype where it will go back, it will like dock itself into a hose and everything coupler and it will then fill itself. So they're they're not there yet, but they're working it. Yeah. They had an autonomous vendor too that goes to it and then middle not this summer at least in the ones I was doing whenever I was offered that they were not on it, but they might be working on it right now. They were out there running to fill it up themselves.
Yeah, Yeah. So on that aspect, legally too, someone has to be in the field identified for safety. So that's a question for Fred that I don't know. I would think that would be the of why he's there. Yeah but I'm saying Yeah but nobody's there to watch it as an air a
2024-10-09