Latest / The Edge Computing Podcast with Fexingo: Local Compute, CDNs, and Distributed Infrastructure / How Edge Computing Powers Precision Agriculture
Transcript
- Lucas: So, when most people think about edge computing, they picture a data center in some warehouse, or maybe a cell tower. But one of the most interesting deployments I've seen recently is actually on a tractor. Luna: A tractor? You mean like, real-time AI processing while it's driving through a field? Lucas: Exactly. John Deere, the agricultural equipment company, has been rolling out edge AI on its newer combine harvesters and sprayers. The idea is simple: instead of sending every image from the camera to the cloud for analysis, the tractor itself runs inference locally to identify weeds in real time. Luna: And you know, it's interesting you mention this because we talk a lot about how edge computing reduces latency and bandwidth costs. In farming, that's not just about convenience—it's about being able to act in the moment. If you're moving at ten miles per hour, you have milliseconds to decide whether to spray a weed or not. Lucas: Right. And that's a huge shift from the old model, where you'd take soil samples, send them off to a lab, wait days for results, then apply fertilizer or herbicide uniformly across the whole field. Now, with edge compute, you can do variable-rate application—spray only where weeds are detected, and only the precise amount. Luna: Which saves money and reduces chemical runoff. But I think the broader point is that this is a case where edge computing isn't just an enabler—it's actually the core requirement. The cloud just can't work here, because the latency is too high. Lucas: Exactly. And if today's tech conversation gave you something usable, something that makes you think differently about how distributed compute is reshaping industries, that's the kind of value we try to deliver every week. The show stays ad-free because of a small group of listeners who chip in at buy me a coffee dot com slash fexingo. No pressure, obviously—just, if you find yourself coming back to these episodes, that's a way to keep them coming. Luna: Yeah, and it genuinely makes a difference. We've heard from folks who use these examples in their own work, or in classroom discussions. So if that resonates, that link is there. Lucas: Alright, back to the field. So John Deere's latest generation of sprayers, the See & Spray Ultimate, uses over 30 cameras and an onboard GPU cluster to process images at a rate of about 1,200 frames per second. The model is trained to distinguish between crop plants and weeds—not just by shape, but by spectral signature. Luna: Spectral signature—so they're looking at how the plant reflects light in different wavelengths? Like, beyond what the human eye can see? Lucas: Exactly. Weeds and crops have slightly different reflectance patterns in the near-infrared range. The edge model picks up on that, and within milliseconds, it triggers a solenoid that opens a nozzle on one specific spray tip. We're talking about targeting a single weed with a droplet of herbicide, while the surrounding crop gets nothing. Luna: That's incredible precision. And it's not just about the hardware—the model itself has to be optimized to run on the edge. You can't just take a cloud-based neural network and drop it onto a tractor. Lucas: Right. John Deere actually had to develop a custom version of their AI model using quantization and pruning to reduce the model size by about 80 percent, while still maintaining over 95 percent accuracy. They also had to handle the variability of field conditions—dust, vibration, changing light throughout the day. Luna: So the edge compute has to be robust, too. That's a theme we see across industrial applications—the hardware has to survive harsh environments, not just server rooms. Lucas: Absolutely. And the results so far are pretty striking. In trials from the 2025 growing season, farmers using this system reported a reduction in herbicide use of about 60 to 70 percent, with no meaningful loss in weed control. That's a huge cost saving and an environmental win. Luna: And I imagine it also reduces the amount of data that needs to be transmitted. If each tractor is generating terabytes of video per day, you can't send that to the cloud over a cellular connection, especially in rural areas with spotty coverage. Lucas: That's a critical point. The FCC's 2026 rural broadband initiative aims to expand coverage, but even then, the latency for cloud processing would be too high for real-time decisions. So edge computing isn't just a nice to have; it's the only viable architecture here. Luna: And this isn't limited to weed detection. The same approach is being used for crop health monitoring, yield estimation, and even autonomous navigation. The tractor can map the field in real time, adjust its route to avoid compaction, and log every spray event for compliance reporting. Lucas: Exactly. And what's interesting is that this is starting to scale beyond John Deere. Smaller ag tech startups like Blue River Technology—which was actually acquired by John Deere a few years back—and others are building similar edge solutions. The components are getting cheaper: the latest NVIDIA Jetson modules can run inference at under 15 watts, which is low enough to be solar-powered on a drone or a stationary field sensor. Luna: So we're moving toward a world where every piece of farm equipment has its own local AI. That's going to generate a lot of data, but it's also going to make farming more efficient and more sustainable. Lucas: And that data doesn't necessarily need to go to the cloud. Some of it can be aggregated at the farm level, maybe on a local edge server in the barn, and only summaries or exceptions get uploaded. That's a hybrid edge-cloud model. Luna: Right. And that's where the network design becomes important. You don't want every tractor phoning home constantly. You want edge nodes that can sync with each other and with a central hub periodically. Lucas: Exactly. Some farms are actually deploying their own private LTE networks or using TV white space spectrum to create a local communications backbone. Edge compute nodes are placed at key locations—maybe at the edge of a field, or on a grain bin—to aggregate data from multiple machines. Luna: So it's not just about the compute on the vehicle; it's about the whole distributed infrastructure. The farm becomes a miniature data center. Lucas: That's a great way to put it. And one company that's been pushing this is a startup called FarmWise—they're actually deploying autonomous weeding robots that run edge AI. These robots are small, electric, and they move slowly through fields, physically removing weeds without any chemicals. Luna: Oh, I've heard of them. They use computer vision to identify each plant and then a tiny robotic arm to pluck the weed. That's even more precise than spraying. Lucas: Right. And again, it's all edge-based. The robot has to make decisions in milliseconds about which plant is a weed and how to remove it, all while navigating uneven terrain. That level of autonomy requires onboard compute that can handle real-time video processing. Luna: So we're seeing a shift from broadcast application—spraying the whole field—to targeted intervention. And edge computing is what makes that targeting possible. Lucas: Exactly. And I think the implications go beyond agriculture. The same principles—local inference, low latency, robustness, bandwidth efficiency—apply to any remote or mobile application. Think of mining trucks, oil rigs, or even disaster response robots. Luna: It's a good reminder that edge computing isn't just about faster Netflix or smarter traffic lights. It's about enabling decision-making in places where the cloud can't reach. Lucas: And in agriculture, that has real economic and environmental impact. We're talking about millions of acres of farmland, billions of dollars in input costs, and significant reductions in chemical use. The edge is literally changing how we grow food. Luna: One thing I'm curious about: what happens to all the data these tractors collect? Is it used for anything beyond the immediate decision? Lucas: Great question. Some of it is used for precision ag analytics—building yield maps, identifying soil variability, planning for next season. That data is typically offloaded at the end of the day via Wi-Fi or cellular, and then processed in the cloud for longer-term insights. But the real-time decisions happen at the edge. Luna: So the edge handles the time-critical stuff, and the cloud handles the retrospective analysis. That's a sensible division of labor. Lucas: Exactly. And as the cost of edge compute continues to drop, we'll see more of these specialized AI models deployed at the edge. The technology is becoming democratized. Luna: I think that's the key takeaway: edge computing is enabling a new class of applications that weren't feasible with cloud-only architectures. And agriculture is just one of many industries that will be transformed. Lucas: Yeah, and we're still early. Most farms today don't have any edge compute—they're still relying on intuition and blanket applications. But the ones that do are seeing measurable returns. I expect that within five years, this will be standard equipment on any new tractor. Luna: That's a bold prediction, but given the pace of innovation, I wouldn't bet against it. Lucas: Alright, that's a wrap for today. Next time, we'll look at how edge computing is being used in autonomous shipping ports—a whole different set of challenges. Until then, thanks for listening.