Published on
Director’s introduction
Welcome to the DAREC Digital Ag Briefing — a weekly, plain-language look at new digital agriculture research and tools, and what they could mean here in Missouri.
This week’s strongest story is not a single gadget. It is a pattern: well-tested camera systems are starting to catch things people cannot watch all day — a sow shifting position in a farrowing pen, a milking step done out of order, or one cow’s movements in a crowded barn. Two of the studies below are especially useful because the researchers checked the camera’s work against careful human observation, instead of only reporting how the software scored on its own test images. That kind of checking is what separates a promising demo from a tool a farm can rely on.
As always, follow the links to the original sources. A summary here is not an endorsement of any product or company.
— Jianfeng Zhou, Director, MU Digital Agriculture Research and Extension Center (DAREC)
At a glance
- In a year-long farrowing trial, sows given long-stemmed straw changed position less often, and piglets were less likely to be crushed in the first three days.
- A camera system in a milking parlor matched human timing measurements almost exactly — and caught hygiene mistakes both human observers missed.
- New tracking software kept individual cows’ identities far more reliably in a crowded barn, though it ran too slowly for real-time alerts.
- In a solar-and-crops field trial in Türkiye, peppers held up well under panels while tomatoes did not — a reminder that results depend on the crop and the place.
- Researchers in Australia are urging caution as farm AI moves from giving advice to taking action on its own.
1. Straw in the farrowing pen, watched by cameras, was linked to fewer crushed piglets
Researchers followed 221 farrowings from 141 sows on a commercial farm over a full year. Sows given long-stemmed straw for nesting had piglet crushing losses of 3.51% in the first three days, compared with 6.52% for sows given a jute bag — the odds of crushing were about 46% lower with straw. Hay did not show the same benefit. Two camera models (YOLOv11) counted how often each sow changed posture, scoring F1 0.92–0.93, and straw sows changed position about 28% less often. Each additional 10 posture changes was linked to about 8% higher crushing odds, which helps explain why the bedding mattered.
What this means for Missouri producers: For Missouri hog producers, the practical point is twofold. First, nesting material is worth testing in your own farrowing rooms — but note this was one farm and one housing system with temporary confinement, so treat it as a reason to trial straw, not a guarantee. Second, the camera method matters beyond this study: using two models and checking where they agree is a sound way to trust automated counts before acting on them.
Source: Frontiers in Animal Science, October 16, 2026 • DOI: 10.3389/fanim.2026.1960568 (Wahmhoff et al.)
2. A milking-parlor camera matched people on timing — and beat them on hygiene checks
A computer-vision system that watches the milking routine was formally validated against human observers on two farms. Its timing measurements — how long teats are prepared before the unit goes on, how long units stay on, and how long cows take to move through — agreed with human measurements almost perfectly (concordance 0.995–0.999). For spotting steps done incorrectly, it was correct every time in this test (100% sensitivity and positive predictive value), and it caught six cases of a towel being reused between cows that both human observers missed.
What this means for Missouri producers: Milking routine consistency affects udder health and milk quality, and nobody can stand in a parlor all day to audit it. A validated camera offers Missouri dairies a continuous check instead of an occasional visit. Keep the limit in mind: validation on two farms is a strong start, not proof it will perform the same way in every parlor layout and lighting condition. Ask any vendor for validation like this — on real farms, against human observers — before paying for accuracy claims.
Source: Frontiers in Animal Science, October 9, 2026 • DOI: 10.3389/fanim.2026.1959684 (University of Guelph validation, two Ontario dairy farms)
3. Tracking every cow in a crowded barn: fewer identity mix-ups, but too slow for live alerts
In free-stall barns where cows block each other from the camera’s view, keeping track of which cow is which is the hard part. On more than 214,000 images, two detection approaches tied on basic detection (mAP50 of 65.4%). The difference was in tracking: the SAM2-based tracker mixed up cow identities only 8–9 times, compared with 50–75 times for the conventional ByteTrack method. The trade-off was speed — the more accurate tracker ran at less than one frame per second — and single images alone could not reliably identify walking or drinking.
What this means for Missouri producers: If the identity is wrong, every health and feeding record attached to it is wrong too, so this accuracy gain matters for Missouri dairy and beef operations considering camera monitoring. For now, read this as a technology to watch for record-keeping and daily summaries, not yet for instant alerts. When a vendor quotes detection accuracy, ask the follow-up question this study highlights: how often does the system lose track of which animal is which?
Source: Smart Agricultural Technology, October 11, 2026 • DOI: 10.1016/j.atech.2026.102622
4. Solar panels and crops on the same land: one system to manage both
A full-season field test in Türkiye combined weather, light, and soil sensors with automated irrigation decisions (using the standard FAO-56 method) and cost tracking over a 122 kWp solar array — an “agrivoltaics” setup where panels and crops share the same ground. Results depended sharply on the crop: pepper yield under panels fell only 1.3%, while tomato yield fell 23–35%. Shade also extended the harvest by about five weeks.
What this means for Missouri producers: Interest in pairing solar with farming is growing in Missouri, and this study is a useful reality check rather than a result to copy directly. The climate, crops, and panel layout in Türkiye differ from Missouri conditions, and one crop thriving while another lost a quarter of its yield shows why crop choice and local trials come first. The transferable idea is the management system itself: sensing, irrigation decisions, and economics tracked together, so a landowner can judge the combined output of food plus power, not either one alone.
Source: Smart Agricultural Technology, October 14, 2026 • DOI: 10.1016/j.atech.2026.102579
5. A field robot that treats disease with light earns a national honor
Saga Robotics’ Thorvald platform, which drives itself through fields at night applying ultraviolet-C (UV-C) light to suppress disease, was named to TIME’s Best Inventions of 2026. According to the report, the system is now protecting about 2,500 acres of California vineyards and roughly 30% of the United Kingdom’s tabletop strawberry production.
What this means for Missouri producers: For Missouri grape, strawberry, and specialty-crop growers, the interest is disease control with light instead of — or alongside — fungicide sprays, which is relevant where resistance, spray timing, or labor are concerns. Treat the acreage figures as company-reported scale, not independent Missouri trial results. Before considering any such service, growers should look for local disease-control data, per-acre cost, and whether the service model (owning a robot versus hiring the treatment) fits their operation.
6. Hog-farm records move to the phone, with paid AI add-ons
South Korea’s WiseLake has brought its PigOS swine management app to iPhone. The app tracks core production measures such as pigs weaned per sow per year, non-productive days, and farrowing rate. Core record-keeping features are free, with AI-assisted analysis offered as paid add-ons.
What this means for Missouri producers: The lesson for Missouri swine producers is less about this specific app — availability, support, and fit for U.S. operations would need checking — and more about the direction of farm software: basic records free or cheap on the phone, with charges for the analysis layered on top. When comparing tools, price the analysis you will actually use, confirm you can export your own data if you leave, and check whose hands your production data passes through.
Source: eFeedLink, October 14, 2026 — South Korea’s WiseLake brings its PigOS farm management app to iPhone
7. Could an eye scan help find tuberculosis in cattle earlier?
Researchers connected to the University of Edinburgh are developing artificial intelligence that looks for signs of bovine tuberculosis (bTB) in images of cattle eyes and retinas. The idea was a runner-up in the Converge KickStart Challenge. The next step is a portable prototype that could be used on farm; accuracy in live animals has not yet been established.
What this means for Missouri producers: File this under “early and worth watching,” not “ready to use.” A quick, non-invasive screening tool would be valuable anywhere cattle health checks are time-consuming, but Missouri producers should wait for published results in live animals and for any diagnostic to go through the normal veterinary and regulatory channels. It is a good example of why this briefing labels the stage of each technology — an award-winning idea and a proven test are different things.
Source: Farmers Guide, October 14, 2026 — University of Edinburgh eye-scanning technology for bovine TB
8. A major new robotics hub for food and farming — in the United Kingdom
The University of Lincoln has won £4.8 million to lead HARVEST, described as the only agri-food hub among eight Innovate UK Robotics Adoption Hubs. The hub includes a £500,000 robot test track and a £1 million Acceleration Fund to help move robotics from laboratories into working food and farm businesses.
What this means for Missouri producers: There is no direct Missouri program here, and the funding is British. The reason to include it is strategic: other countries are investing in shared test facilities and adoption support, not just research, to get robots onto real farms faster. That is the same gap DAREC works on — testing technology under Missouri conditions and helping producers judge it — and this hub will be a useful benchmark for what adoption support can look like.
Source: October 13, 2026 — University of Lincoln wins £4.8 million for the HARVEST agri-food robotics hub
9. As farm AI starts acting on its own, researchers urge hard limits
Australia’s national science agency, CSIRO, and other experts are warning about “AI agents” — software that moves beyond giving advice to taking actions, such as ordering, scheduling, or operating connected systems. They note that a misread record or a manipulated instruction could turn a small error into a costly one. Their recommended safeguards are practical: keep systems read-only by default, let the AI draft an action rather than do it, set hard limits, keep a manual override, and log everything the system does.
What this means for Missouri producers: This applies directly to Missouri farms as AI features appear inside equipment, feeding, irrigation, and record systems you already use. Before enabling any feature that acts without asking, check four things: what it is allowed to do on its own, what spending or operating limit is set, how you override it, and whether you can see a log of what it did. Advice you can ignore is low-risk; action you cannot undo deserves a much higher bar.
Source: October 12, 2026 — CSIRO and others on AI agents on Australian farms
A note on sources
Each item above summarizes the linked report or journal article as published on the date shown. Figures for commercial products are reported as the company or outlet stated them and have not been independently tested by DAREC or the University of Missouri. Mention in this briefing is for information only and is not an endorsement.
Questions or a technology you would like DAREC to look into? Contact the center through darec.missouri.edu. Email: digitalag@missouri.edu.