Right now, agriculture is in the hands of multiple individuals, making it disorganized. It has not been rolled up under big corporations and there are still a lot of independent business people in some respects. What is happening now with ag tech startups is there are tremendous amounts of venture capital going into ag technology as more people realize that data is going to start to consolidate agriculture.
There are a number of technological advancements coming up that are going to change the agriculture business. I used to talk about the future being 15-20 years out. Then I started to talk about it being 5-10 years out. Now I’m talking about it being a few years out. These technological advancements are already out there, working in another industry or they are being developed right now.
At Ohio State, my PhD student Chris Weigman has been working on the latest in artificial intelligence, particularly in neural network classifiers. Neural networks are essentially a computer representation of what is going on in the human brain. What Chris is working on is training these neural network classifiers to recognize issues in crops like nitrogen deficiency.
Technology to Keep an Eye On
In this evolving technological world, there are a number of advancements worth watching in the agriculture world:
- Amazon Web Services (AWS) is building cloud storage facilities and purchasing loads of file servers. When we buy a file server, we get one with all the bells and whistles which is what leaves us open to nefarious attacks. In only getting certain features on their servers, AWS is better able to protect their file servers.
- Autonomous Tractors are being introduced on a regular basis and I encourage people to think about them in terms of how they will change how dealerships run. If you go a step further and eliminate the human operator, you are able to cheapen the environment because you do not need the cab or the seating.
- Cryptocurrency allows for money to take place in a transaction over the internet. Currently, you do everything with wire transfers and banks so once the government backs this, it will be a game changer.
- Block Chain is when you have data that is encrypted and you break it up, putting it into a bunch of different file servers so it is harder for nefarious people to take advantage of the situation. The great thing about block chain is that the fund transfer takes place in microseconds and the transactions can be tracked back to its origin.
Take, for instance, teaching a child to pour a glass of milk. It is about tipping the carton over, having the glass in the right position and making sure the milk makes it into the glass. Most children learn to do this, but it often takes a couple tries. If you think about it, this is human learning and what we’re trying to do is emulate that in computers today.
The human brain has connected nerve cells and neural networks have neural net nodes. Each of these nodes have input and output links. The input link has a weight associated with it and these weights are modified to bring the network’s output to the goal behavior.
While there is more than one kind of neural network, the most important to our work are convolution neural networks (CNNs). With these, we take an image of a plant that has a disease and we do a convolution of that image. We take a small subset or mask out of the image and do another convolution of that, going across until we get to the final layer of nodes which is where we get an output.
The best way to think about training neural network classifiers is in comparison to crop scouts. Most crop scouts do not get much more than 50 feet into the field and they usually go in off of a county or farm road. This means the crop scout is generally looking at about 10-15% of the field and making management decisions based on that small percentile.
Neural networks are a computer representation of what is going on in the human brain. What we are doing is trying to emulate human learning in computers.
Training Neural Network Classifiers
To train neural network classifiers, we first started by collecting images of fields. Using a multi-rotor drone, we fly it over the field and take 300-400 images. We look at the center of the image and that gives us a good indication of what is going on in the field.
At the end of the pendant on the drone is a camera head that is dropped down into the plant canopy. Last summer, Chris took photos of soybean plants with different problems in the field; sudden death syndrome, frog eye leaf spot and dicamba damage. We later went and did the same thing in corn fields, looking for both healthy and diseased plants.
So, how do you give a computer the ability to distinguish between those diseases?
Chris again went out into the field and collected images of corns plants, both healthy and diseased. He took thousands of images and ran them through a program for several days. Within 4.5 days, we had trained the neural network classifier to recognize those images. At this point, Chris has got around 93% classification accuracy.
Now, is this better than a human? Our goal with this is to leave the farm with a map that says, “Do this to your corn field, do this to your soybean field and you’ll make money doing it.”
We are using a multi-rotor drone with a camera at the end that we drop done into the plant canopy.
We are nearly to the point where we can get the neural network classifier trained enough to be placed on the drone and let the drone indicate what is occurring in the field. Some of this training is memorization, some of it is experience but it is all those experiences that come to reality.
There is a company — Blue River Technology — using similar technology to what we are working on. The company’s first major product, the lettuce bot, uses artificial intelligence to identify and count lettuce plants in the field. It then accurately sprays a herbicide to kill the unwanted plants. This technology has been expanded on to control weeds in Roundup Ready cotton.
What it Means for the Future
Everything is internet connected today and many are connected to more than one device. Think about all the devices in the field. Our data ecosystems are continuously evolving, and these new technologies are changing the scope of agriculture.
The animal industry is already using a lot of this advanced technology. There are multi-thousand cow operations that are doing well right now and many are using robotic milkers and feeding equipment. The farmers themselves do not have to be on the property because if something goes wrong, their iPad will notify them.
There are robotic feed mixers that mix 7 or 8 dairy rations, traveling around the farm to feed dairy cows. Even the manure is scooped robotically.
These technologies and changes are not 5 or 10 years down the road; technology on farms is happening today. There are going to be a lot of things that come along, and some will be snake oil while others will perform. In today’s world, you have to adapt to technology to stay in business.