Remove Biases From Food Quality Inspections With AI

Miku Jha, Founder and CEO of AgShift, started thinking about "closing the gap between food and technology" in 2016 when California was facing a drought. Her solution to the food waste problem, particularly in the supply chain, is a digitized inspector called Hydra F100 BQ. This analyzer uses artificial intelligence to assess fruits’ quality, which provides an "unbiased pair of eyes," speeds the inspection process and reduces food wastage.

By Pragati Verma, Contributor

At a Central California-based cooling facility owned by Driscoll’s, one of the largest fresh multi-berry producer in the world, food quality analyzers inspect berries every day, assessing the fruit’s weight, size, color, scouring for defects like mold and bruises, and ensuring the fruit meets United States Department of Agriculture (USDA) and other customized specifications. These particular analyzers helping food inspectors are machines, designed and developed by Sunnyvale-based AgShift.

Called Hydra F100 BQ, the analyzer uses computer vision and deep learning algorithms to assess fruits’ quality and autograde the fruit as a whole. The digitized inspection process, currently in commercial trial, is working well for Driscoll’s.

“AgShift provides a promising technology that can alleviate many complex issues related to quality assessment of berries for us,” says Brie Reiter Smith, head of quality at Driscoll’s. “We are happy with the progress that this young company has made in a short span of time.”

“AgShift provides a promising technology that can alleviate many complex issues related to quality assessment of berries for us.”

—Brie Reiter Smith, Head of Quality, Driscoll’s

According to AgShift founder and CEO Miku Jha, Hydra is aiming to augment food quality inspectors. For one, the job of counting the number of blueberries in a clamshell is “painstaking and leads to mental and physical fatigue over time.”

“In addition, there are a lot of visual defects that are not only straining but also next to impossible to judge with human eyes,” she adds, “Powered by artificial intelligence, Hydra makes an unbiased, objective prediction and serves as a powerful digital assistant to the inspectors.”

Hydra, however, is not their only product. AgShift has also developed AgSkore, a mobile-based inspection workflow digitization solution, powered by analytics in the backend. Both the products, she says, tie to the bigger vision of AgShift, which is to “bring better insights and transparency on food quality from farm to fork.”

Consistent ‘Eyes’

Jha says AgShift’s analyzer improves food quality assessment in three ways: “The current food quality assessment processes are paper-based and manual, many times leading to inconsistent and subjective outcomes. For example, what’s red to you might be pink to me,” she explains. Their analyzer, Jha insists, is “one of the first tools with uniform, unbiased pair of eyes, so to speak, to objectively and consistently assess the quality of commodities.”

AgShift’s analyzer improves food quality assessment in three ways:
1. objectively and consistently assessing food quality with an “unbiased pair of eyes”
2. speeding the inspection process from 6-8 minutes to a mere 20 seconds; and
3. reducing food wastage

Secondly, it makes the process much faster. “A typical strawberry clamshell contains 40 to 50 berries and inspectors visually eyeball and inspect four to five boxes. The process takes six to eight minutes,” she estimates, “Hydra can inspect a sample and produce a grade report in a mere 20 seconds.” A faster process will obviously reduce labor costs and improve operational savings. According to her, producers can save time or use the faster process to inspect bigger samples in the same time and “have more data points to assess quality.”

Thirdly, it can help fix food loss and wastage. “Everyone talks about how consumers toss away food, but food is also wasted throughout the supply chain from initial agricultural production to the local store.” Jha expects standardized food inspection across the entire supply chain to reduce food wastage—1.3 billion tons of food or about a third of what we grow every year.

Mission to Cut Food Waste

This quest to cut food waste is what initially inspired Jha to foray into agriculture. A technologist who grew up on a mango farm in India, she started thinking of “closing the gap between food and technology back in 2016 when California was facing a drought.” She began by driving to various farming communities to identify their biggest challenges and became interested in solving the food waste problem, especially in the supply chain.

She found that inconsistent inspections or quality assessments lead to rejection of food. “We decided to focus on empowering food quality inspectors in the food industry with the ability to provide fast and consistent food quality assessment that for decades has been relegated to manual processes with cumbersome paperwork,” she recalls.

Today, AgShift has raised $5 million in seed round from investors like Exfinity Ventures and CerraCap Ventures, and its analyzers are in commercial trials. Bringing operational efficiencies and objective accuracies across the entire food supply chain, however, was not easy. According to Jha, one of the biggest challenges was the lack of a well-curated and labeled data sets in the food industry, and their deep learning model needs extensive real-world images to train the software on different attributes. She elaborates with an example: “To identify decay or mold in strawberries, our software first needs to understand what decay or mold looks like. We need to feed a large number of images of strawberries with decay or mold for it to understand what is decay and what is mold.”

Jha says AgShift has created its own extensive data set for the commodities they are working on, and the software is now ready to use. “And for the rest of the commodities, we now have our own way of augmenting it from a small amount of seed data set,” she adds.

Ripe for Commercialization

According to Jha, Hydra has achieved high levels of prediction accuracies in inspection of commodities. At Driscoll’s, she expects to “move from last stage of commercial trials to production deployment at scale by the end of the year.” Smith seems to agree. “As part of trials,” she says, “we are currently validating the system results and will be looking to commercialize the solution in our selected facilities once it meets our product acceptance criteria.”

AgShift’s analyzers are not only running at Driscoll’s—or inspecting just produce for that matter. Olam, one of the world’s largest agriculture commodity companies, is using them to inspect cashews at its processing facility in Vietnam, for instance.

Jha plans to expand to seafood next, as “it has high perishability and waste.” AgShift’s deep learning models, she adds, can help the seafood supply chain with “objective species identification and freshness assessment, as well as determine any change in quality of fish as they move through the supply chain.”

The food industry is ripe for innovation, Jha believes. “[The] openness to technology innovation is encouraging people like us, who are on a mission to tackle complex, global challenges facing the food ecosystem.”