by Merritt Jenkins, Summer Associate 2020

At LDV Capital, we invest in people building businesses powered by visual technology. We thrive on collaborating with deep tech teams who leverage computer vision, machine learning and artificial intelligence to analyze visual data. Every summer we complete in-depth research for our annual LDV insights reports. These reports are a deep dive into a specific industry and the visual technologies that we believe will be critical to the future of that industry. 

In our latest LDV Insights Report: “Visual Tech Driving Innovation in 15 Food & Agriculture Sectors”, we examined the opportunities for visual technologies across the meat supply chain. From improved animal quality of life to efficient harvest, there are exciting new applications for computer vision and machine learning. Merritt Jenkins, Summer Associate at LDV in 2020, presents some of our findings.

The meat industry traditionally has not received significant interest from venture capital, which is exactly why I chose to research the industry for our Food & Agriculture Insights Report. I have spent the past five years of my career focused on robotics and computer vision in agriculture, and I have often hypothesized that animal agriculture is an untapped frontier in precision agtech. While billions of dollars poured into indoor farming, crop gene editing, and farm management software last year, only a small fraction was invested in livestock technology. After speaking with many experts in the space, it is clear that the role of computer vision in meat processing facilities and animal feeding operations will grow tremendously.

Computer vision will play a large role in meat processing facilities

“There are many opportunities for operational efficiency improvements in the meat industry”, says Decker Walker, Partner and Managing Director of The Boston Consulting Group’s Industrial Goods practice, “but the greatest opportunity for computer vision and robotics is in slaughter and primary processing.” These processing facilities rely on manual labor for steps such as inspection, cutting, and deboning, all of which will be automated in the future.

Inspection offers the lowest-hanging opportunity in processing facilities because cameras can supplement or replace a process traditionally performed by a plant employee. Poultry equipment suppliers such as Marel have developed camera systems to inspect birds at the defeathering stage, and P&P Optica is pioneering the use of hyperspectral imaging to detect contamination, foreign objects, and bones in the meat. Tyson is deploying machine learning-enabled vision systems this January to inspect beef carcasses, and eventually plans on fully automating the role of the FDA inspector. It is entirely feasible that inspection processes will be dominated by cameras within the next 5-10 years.

Primal cuts will likely be the second phase of automation in processing facilities. This process consists of breaking large carcasses down into small pieces and is traditionally performed manually with powered saws. Automating these cuts can both improve yield and employee safety. Most primal cut systems rely on 3D scanning with structured light to identify external geometry, but some systems also utilize dual-energy x-ray absorptiometry (DEXA) to detect bones, fat, and muscle to improve the cut location. Facilities in Europe and Australia/New Zealand, such as Scott Automation’s lamb processing facilities or Denmark’s Danish Crown swine slaughterhouse, have already automated primal cuts, but the process remains largely manual in the U.S. The lack of primal cut automation in the U.S. is due to lower wages and meat prices, but this is likely to change as companies such as Tyson and JBS are motivated by COVID-19 to explore ways to reduce headcount in their processing facilities.

One of the last frontiers in meat processing automation will be deboning, which is highly manual and requires the highest concentration of workers. Automated deboning machines currently cannot compete on throughput or cost with the well-trained hands of manual workers, but companies such as Tyson and JBS are investing significant capital into robotics R&D. As the price of x-ray and hyperspectral imaging decreases and machine learning algorithms improve, we may see more cost-competitive automated deboning systems. 

In summary, computer vision and robotics will make inroads at different rates within different stages of meat processing. Vision-enabled inspection will likely see the quickest uptake because camera systems take up no additional floor space and are an item of relatively small capital expenditure. Primal cuts may be automated next because of improved yield and worker safety, and deboning will be the last frontier due to dexterous manipulation challenges. Computer vision offers the opportunity to improve worker productivity and will play an increased role in meat processing facilities over the next decade.

Cameras on the farm will improve feed efficiency and animal quality of life

If meat processing facilities have been slow to adopt computer vision, then feedlots and barns are in the dark ages. This is not without good reason, however. In much of the rural United States, broadband internet is nonexistent or slow, and field conditions such as dust pose challenges to cameras. Yet there is an opportunity to streamline this highly manual industry. Technology startups are beginning to address issues such as access to feed, livestock health, and harvest selection.

A universal challenge in livestock production is ensuring that the animals have constant access to feed. The dairy industry is leading innovation in feed delivery and measurement: Lely has sold over 10,000 feed pusher robots and Cainthus deploys fixed cameras to measure feed access. Similar technology has yet to be developed for cattle, pigs, and chickens, but it’s on the horizon.

Livestock health is a big business, and the first defense in animal care is a well-trained human eye. For example, in cattle feedlots “pen riders” on horseback evaluate cattle health from afar. In swine and poultry barns, farmers and veterinarians evaluate animal health by foot, which risks introducing disease when vets travel from farm to farm. Cameras in the feedlot and barn will supplement the need for a manual health evaluation, and veterinarians may use camera footage to better diagnose illness.

When the time comes for harvest, animals are frequently selected manually. Individual pigs are visually identified by a well-trained technician, and chickens are manually weighed before loading and transporting. Vision-based estimation of animal mass has been an active area of academic research for the past decade and limitations such as computation cost and algorithm accuracy are beginning to lift.

The feedlots and barns of the future will rely on cameras to free up labor and improve production efficiency. Livestock will be evaluated by vision algorithms as they consume food, ensuring that they have consistent access to feed and are acting healthy. If something appears to be wrong with an animal, a veterinarian may be able to diagnose the issue remotely. And when animals reach the desired size, the farmer knows in advance which ones will be taken to the processing facility that day.

Unlike meat processing, the rate and order of technology adoption in feedlots and barns is hazy. Yet the opportunity is clear: continuous monitoring using computer vision will boost feed conversion efficiency, improve animal health, and reduce labor needs.

Visual technologies in animal agriculture face headwinds, but are exciting opportunities

While opportunities abound, there are headwinds too. Some challenges are specific to the animal agriculture industry, such as industry regulations or technical challenges to developing livestock identification algorithms. Cainthus, a well-known startup in the dairy industry, initially focused on “facial recognition for cows” but later pivoted to feed management due to algorithm development challenges. 

Other challenges are endemic to the general agtech industry: lack of broadband internet in swine and poultry barns constrains opportunities for connected cameras; product development cycles are held back by the speed at which an animal grows; and quantification of results is restricted by the lack of upstream and downstream data collection. Another challenge is that, despite the significant scale of the livestock industry, very few singular tasks are costly enough to justify venture capital investment. The successful companies will manage to solve several problems at once, such as swine health and harvest readiness.

Many of today’s headwinds will be tomorrow’s competitive moats, and the macroeconomic trends are compelling. Labor is increasingly difficult to find in rural areas, Americans are newly aware of health and safety issues in meat processing facilities, and customers want increased supply chain transparency. The U.S. meat and poultry industry accounts for $1 trillion of total economic output, and we at LDV are taking a close look at startups innovating in the space. If you’re building a company in this sector, we want to hear from you. We are always looking to partner and invest in brilliant people.