Editor's View: AI Can Spot A Blemish But Fresh Produce Still Needs Human Judgement

Anyone who has spent time around fresh produce, flowers or plants knows that visual inspection is not as straightforward as it sounds.

On paper, it is simple enough. Check the colour. Look for damage. Assess size, shape, uniformity, decay, presentation and shelf life. Decide whether the product meets the spec.

In reality, it is rarely that neat.

A punnet of strawberries can look perfectly acceptable from one angle and disappointing from another. A bunch of roses may appear strong at first glance, until petal bruising, a bent head or the early signs of botrytis become obvious. A tray of young plants may be healthy and saleable, but uneven enough to cause debate over whether it meets a customer’s expectations.

This is the space where technology is now making a serious move.

Artificial intelligence, computer vision, cameras and sensors are increasingly being used to support quality assessment across fresh produce and horticulture. These systems can examine images, detect surface defects, assess colour and shape, flag inconsistencies and help grading teams make faster, more consistent decisions.

There is no doubt that the technology is advancing quickly.

In the United States, Albertsons Companies announced in May 2026 that it had introduced an AI-powered Intelligent Quality Control tool for fresh produce, built with Google Cloud technology. The system is designed to support distribution centre inspectors by analysing produce images against company quality standards. It has initially focused on strawberries and red and green grapes, with plans to expand to other fruits.

Elsewhere, post-harvest sorting technology is already being used to identify defects in crops such as citrus. TOMRA Food’s LUCAi deep-learning system, used with its Spectrim platform, is promoted for detecting a wide range of citrus defects, including rot, sunburn and clipper cut.

The academic work is moving in the same direction. Research into AI and food quality has shown that vision-based fruit sorting can achieve high levels of accuracy in grading, although researchers are also clear that standard image-based systems mostly deal with what can be seen on the surface. To understand internal quality, businesses may need other technologies, such as hyperspectral imaging or X-ray systems.

That distinction matters.

Fresh produce quality is not just a visual exercise. A product can look good and still disappoint on firmness, flavour, shelf life or internal condition. Equally, a product can have a small cosmetic defect and still be perfectly edible, saleable and commercially sensible to use.

This is where the conversation becomes more interesting — and more important.

AI may be very good at spotting what it has been trained to recognise. It may identify blemishes, bruising, discolouration, misshapen product or uneven grading more consistently than tired human eyes at the end of a long shift. It may also produce a useful image record, which could help settle disputes, improve supplier conversations and build a clearer picture of recurring quality issues.

But it does not automatically understand the full commercial context.

A small mark on a citrus fruit, a leaf tear on a plant, a slightly open flower or a colour variation in a grape does not mean the same thing in every situation. It depends on the customer, the specification, the destination, the price point, the season, the available supply and the likely shelf life.

That is not a simple yes-or-no decision. That is judgement.

The same applies to cut flowers and plants. Floriculture has its own inspection challenges: petal damage, bent stems, uneven opening, leaf quality, flower count, height, colour, bud stage, disease symptoms and overall presentation. Researchers are already exploring deep-learning methods for detecting damage in cut roses and monitoring vase life and grey mould. Plant grading systems using high-resolution camera technology are also being used to assess features such as height, volume, colour and number of flowers.

For a sector dealing with highly variable, living products, the appeal is obvious. Visual inspection has always depended heavily on skilled people making quick decisions under pressure. Those people bring experience, but consistency can vary between sites, shifts, customers and seasons. One inspector may pass what another rejects. One market may tolerate what another will not.

AI offers the possibility of greater consistency.

It does not get tired. It does not have a bad day. It does not become more lenient because the intake window is closing or more cautious because a customer complaint came in the week before. It can apply the same logic repeatedly and produce a data trail behind the decision.

That could be genuinely useful.

If businesses can capture better quality data, they can start to see patterns more clearly. Which suppliers are consistently performing well? Which varieties are vulnerable to damage? Where is the cold chain weakening? Are defects arising at harvest, in transit, in storage or during handling? Are products being rejected because of genuine quality concerns, or because the specification is being interpreted differently at different points in the chain?

Used well, AI could help reduce waste as well as improve quality.

That is a prize worth chasing. The fresh produce sector already works with tight margins, short shelf lives and significant pressure around availability, labour and compliance. If better inspection can help more product find the right route to market, rather than being rejected, downgraded or reworked unnecessarily, then it deserves serious attention.

But this is not the moment for blind enthusiasm.

Technology suppliers often talk about AI as if it can see everything, know everything and fix everything. It can't!

Computer vision depends on good images, good lighting, strong training data and clear rules. It can be affected by condensation, glare, mud, overlapping product, packaging, unusual defects and natural variation. A system trained on one crop, one variety or one set of standards may not transfer neatly to another. A model that works well in a controlled environment may struggle in a busy, messy, real-world packhouse.

And even when the system works, businesses still need to ask what happens next.

Who reviews the decision? How are borderline cases handled? Can the system explain why it has flagged a defect? What happens if it rejects too much product? What happens if it misses something serious? Who owns the images and the quality data? Can suppliers challenge the outcome? Does the system support fairer decisions, or simply make rejection faster?

These are not minor details. They go to the heart of trust.

The most sensible future is not one where AI replaces inspectors, technical managers, growers or packhouse teams. The best future is one where it supports them.

The industry does not need technology that pretends nature is uniform. It needs tools that understand, or at least can help manage, the fact that it is not. Fresh produce, flowers and plants are living products. They vary by season, variety, growing conditions, handling and time. No two items are identical, however much the supply chain sometimes wishes they were.

That is why human expertise remains so important.

A good inspector does not just see a defect. They understand what it means. They know when a mark is cosmetic and when it signals a bigger problem. They know when a conversation with a customer is needed, when product can be reworked, when it should be downgraded and when it genuinely should not go any further.

AI can sharpen that process. It can provide evidence, consistency and speed. It can help people make better decisions. It can turn visual inspection from a hurried judgement call into a more transparent, data-led process.

But it should not be allowed to become a blunt instrument.

The fresh produce, cut flower and plant sectors should absolutely explore AI inspection. They should trial it, challenge it, improve it and use it where it adds value. But they should do so with their eyes open.

The right question is not whether AI can spot a blemish. The better question is whether it can help the industry make fairer, faster and more commercially sensible decisions about real products in real conditions. That is where the opportunity lies.

Because in this industry, quality has never been just about what you can see. It is about what you understand.