The Growing Impact of Machine Vision in Modern Agriculture

The agriculture industry faces mounting pressures from a growing global population, unpredictable weather patterns brought on by climate changes, and increasing demand for sustainable practices. In this environment, machine vision has an opportunity to transform the industry.

Recent advances in machine vision technologies — such as deep learning, hyperspectral imaging, and real-time processing — have opened new doors for precision agriculture. These innovations are being harnessed in various applications, including crop monitoring, automated harvesting, weed detection, and livestock management. Whether identifying diseased plants before they spread or ensuring that crops are harvested at peak ripeness, machine vision enhances both productivity and sustainability.

The path to full-scale adoption is not without challenges, however. High implementation costs, the need for skilled operators — and even just the natural variability in agricultural environments and output — pose significant hurdles. Despite these obstacles, ongoing research and investment move us toward a future where machine vision becomes an integral part of agriculture.

Machine Vision’s Capabilities

Most advanced farming operations are at least familiar with the proficiencies of machine vision in the agriculture industry, says David Dechow, founder and owner of Machine Vision Source, a machine vision integrator.

But adoption varies significantly across regions and farming types, says Minna Törmälä, global marketing manager for Specim, which specializes in hyperspectral imaging systems. Industrial-scale farms and vertical farming tend to lead the way. “There is still significant educational work to do — particularly for smaller-scale and traditional farms — to understand how machine vision can improve profitability and sustainability,” she says. 

Applications and their benefits run the gamut across the agricultural landscape, Törmälä points out. In the field, machine vision can be used to monitor crops to detect nutrient deficiencies, diseases, or pest infestations early, and provide a better understanding of where to apply fertilization or pesticides. It can assess plant health and maturity to estimate yields more accurately and plan harvests accordingly. Soil and water analysis using machine vision can help optimize irrigation and soil amendments.

“Vision has been deployed for some time in the execution of precision agriculture,” Dechow says. “Drones with spectral or hyperspectral imaging scan fields to detect areas of slow growth, disease, and low water availability or consumption.”

Postharvest, machine vision plays an important role in sorting and grading by providing automated inspection for safety and quality standards. “Besides detecting contaminants and defects, hyperspectral imaging can identify spoilage, bruising, internal defects, and measure even taste in fresh fruits and vegetables,” Törmälä notes.

In addition to grading and defect detection, some of the more mature uses of machine vision include guidance for milking operations and quality evaluation in egg production and packaging, Dechow notes.

The Benefits to Be Reaped

One of the biggest reasons farming operations are turning to machine vision is the inability to get enough farm labor to otherwise get the work done. Automation can reduce labor costs, particularly at larger scales. But in many cases, operations simply can’t find the workers.

Nature Fresh Farms, based in Ontario, Canada, is one of North America’s largest greenhouse farms, growing organic berries, peppers, tomatoes, and cucumbers. The company turned to robotics and machine vision for its harvesting operations largely because of its struggle to find labor, according to Cornelius Neufeld, executive vice president for Nature Fresh.

“Our biggest challenge is labor,” he says in a video case study with FANUC and system integrator Four Growers. “As this sector grows more and more and more, we really need more automation to do all the jobs. We just don’t have enough people in this industry anymore.”

But labor issues are certainly not the only reason to turn to machine vision in farming. Imaging techniques can address critical challenges in crop monitoring, yield optimization, and resource efficiency. “The agriculture industry increasingly recognizes machine vision’s potential for providing detailed insights into plant health, soil conditions, pest detection, and post-harvest quality control,” Törmälä says.

Healthier crops, better use of precious resources, and consistent quality are key benefits of using machine vision in agriculture, she adds.

Applications Down on the Farm

Dechow has been involved in a number of traditional agriculture projects, some of which involve sortation and damage detection of eggs prior to packaging and robotic handling of fruits and vegetables (including apples and brussels sprouts) for automated cutting and trimming.

Not all applications yield good results, however, he says. “That’s not necessarily due to limitations in machine vision alone,” he says. “In several cases, the capability of mechanical automation becomes the main challenge, particularly with respect to speed of process, reliability, and cost. In robotic handling, for example, gripping technology has sometimes been an important obstacle.”

In general, in-field vision-guided robot harvesting for fruits and vegetables is not particularly ready for prime time, Dechow contends. Vision still faces some limitations in the field — as do automation and gripping. “Overall, too, the speed of processing using a robot is, in most cases, way too slow for the system to attain a return on investment,” he adds.

Success in a Controlled Environment

Where machine vision is finding more success so far is largely in vertical farming, where more controlled environments provide the consistency needed.

There’s already a push toward controlled environment agriculture, according to Interact Analysis, to counteract various factors brought on by climate change, such as reduced water availability and degradation of topsoil, at a time when we have an increasing number of mouths to feed.

“Controlled environment agriculture enables a more efficient use of key resources,” notes Blake Griffin, research manager for Interact Analysis. “Using carefully controlled water, air, and lighting systems, these growing environments can produce food at a consistent rate and can more easily eliminate pests and disease from plants.”

Such environments also help create the right surroundings for machine vision applications, Dechow notes. “Applications in vertical farming have specific potential in that the product is very constrained while growing,” he says. “Machine vision-enabled planting and harvesting are strong value-added solutions for this industry, and the more organized structures for growing in vertical farming lend themselves to more reliable automation.”

As a system integrator, Dechow has put together systems in both standard agriculture and vertical farming. In vertical farming, one client was able to achieve a significant time reduction to plant seedlings into growing towers. “This was accomplished using vision-guided robotics to locate the seedlings and the tower receptacles,” he says.

The application was achieved with FANUC robots and a custom machine vision solution. “The critical gripping mechanism was designed by the customer and was specific to the plant being processed,” Dechow says.

ROI in this case was achieved not only because of the reduced planting time but also because of planting repeatability and the scalability over a broad range of planting processes. Because vertical farming faces a finite space for the plants to grow, the speed of replanting is a considerable factor in throughput, Dechow points out.

Vertical farming and controlled environment agriculture tend to be at the forefront of machine vision adoption, Törmälä agrees, relying heavily on precision technologies to maximize yield per square meter.

“However, open-field farming is also making strides, especially in regions where precision agriculture is a growing focus,” she says. “Key areas include high-value crops like fruits, vegetables, and vineyards, where even minor quality improvements significantly impact revenue.”

Postharvest Sorting and Grading

Examples of these high-value crops include avocados, blueberries, strawberries, and tomatoes. In one such case, Dutch company Condi Food used hyperspectral imaging to develop an innovative inspection system for evaluating the taste of cherry tomatoes in real time.

Traditional techniques sort tomatoes based on size and color, but that’s an imperfect way to judge taste. Using a Specim FX17 near-infrared hyperspectral camera, Condi Food efficiently and nondestructively scans cherry tomatoes to analyze their spectral signature. This provides important information about key taste attributes, including ripeness, Brix, and acidity.

Condi Food developed a taste model that uses the hyperspectral data to analyze the elements and characteristics relevant to taste. This type of model could be used for a wide range of fruits and vegetables beyond tomatoes. In this application, the Specim camera enables detection at production speeds of up to 60 cherry tomatoes per second. 

Promising Work in New Applications
Although newer, emerging use cases have been well publicized, Dechow says, few have been widely adopted. “In many cases, this is due to the systems not being able to fully achieve their stated purpose and reliability,” he says.

As an example, Dechow points to the use of 3D vision-guided robots for discrete fruit picking. Apples and tomatoes, for instance, can cause issues simply because of their variability. “The technological obstacles have been and remain the inconsistencies in the natural product and its growth,” he says. “A related consideration is speed of operation, in particular for harvesting, where automated systems may not achieve required rates of production.”

Advancements in machine vision technologies should continue to help promote their use in agriculture. “Smaller, higher-resolution, and lower-cost imaging systems in 2D, color, and 3D — as well as improvements to multispectral and hyperspectral imaging — may help agricultural solutions development,” Dechow says. “Cost being a big factor, ongoing component cost reductions as may be seen in 3D cameras will be beneficial to agricultural systems.”

Artificial intelligence (AI) could lend a considerable hand as machine vision takes on the less predictable world of farming. “In particular, the ability of deep learning segmentation is well suited to overcome the inherent variations of natural products and facilitates more reliable object location in agricultural applications,” Dechow says.

In the Nature Fresh Farms example, Four Growers developed an AI system that enables the vision to provide deeper analysis of the harvesting environment.

The system captures images from multiple stereo cameras, stitching them together in a 3D space to create a cohesive scene. “From there, our AI is able to go through and understand not just tomatoes and not just stems but actually understand different concepts of the scene,” says Brandon Contino, CEO of Four Growers. “It uses that information to then figure out what is the best way to approach each individual tomato to successfully pick it as well as what is the most efficient path to take.”

Data analysis will be a key aspect to AI’s ability to improve hyperspectral imaging as well, Törmälä notes. “AI will allow more efficient data analysis and real-time decision-making in the field — for instance, cameras capable of processing spectral data directly onboard,” she says.

Commercial Considerations for Machine Vision Success

Some commercial considerations that strongly impact adoption of vision systems for agricultural use include the cost and complexity of the systems, as well as the processing environments, Dechow says.

In hyperspectral imaging, Törmälä says, one of the main challenges is the complexity of interpreting spectral data, especially for users without prior experience. Specim is trying to address this in part by providing more user-friendly solutions with more portability and on-site analytics.

“The adoption of hyperspectral technology in agriculture is currently more feasible for larger companies with the necessary resources and expertise, as its complexity requires a level of knowledge not typically expected from individual farmers,” Törmälä says. “To expand its reach, there is a growing need for accessible services and turnkey solutions that simplify its implementation and usage.”

Hyperspectral imaging is well known for its ability to detect information that is invisible to the human eye, as with the taste example from Condi Food. However, the complexity of data acquisition and analysis has hindered its widespread adoption.

Using Specim’s hyperspectral cameras, Wageningen University & Research (WUR) in the Netherlands developed a smart, easy-to-use laboratory system for reliable analysis of fresh produce. The system enables users without in-depth knowledge of hyperspectral imaging to precisely analyze moisture and soluble solids content in a range of fresh fruits.

Universities and research institutes will continue to play an important role in bridging the gap between technology providers and end users by advancing knowledge and developing applications, Törmälä adds.

“Automation clearly could improve production if it is reliable and supportable and if the ROI can justify the equipment cost,” Dechow says.