Apple Colour Assessment with AI: Ripeness Checks, Starch-Iodine Tests, Leaf Area, Leaf Greenness and Disease Spots

Why does apple colour matter for ripeness?

Apple colour matters because it reflects two different processes happening in the skin, and only one of them is closely tied to maturity.

1. Background colour (ground colour). This is the underlying colour of the skin, beneath any red. As the fruit matures, chlorophyll breaks down and the background shifts from green towards yellow. Because chlorophyll loss is part of the ripening process, background colour is the more reliable colour-based maturity indicator.
2. Blush (over-colour). This is the red, produced by anthocyanin pigments. Blush development depends strongly on sunlight reaching the fruit, on cool nights, on the variety and on the position of the fruit in the canopy. A well-exposed apple can be deeply red while still immature, and a shaded apple from the same tree can be ready but pale.

That is why experienced growers and advisers look at the background colour on the shaded side of the fruit, not just the red side, and why colour is always read alongside other maturity tests.

Colour also matters commercially, regardless of ripeness. Many supply specifications set a minimum percentage of red over-colour for a class or grade, so blush coverage directly affects packout and price.

What does colour not tell you?

Colour alone does not measure firmness, sugar content or starch conversion. A complete maturity picture usually combines:

AI does not replace this list. It makes the visual items on it — colour and the starch-iodine test — measurable and repeatable, and it makes recording much faster.

How does AI measure apple colour from a photo?

AI measures apple colour by first finding each fruit in the image, then outlining it precisely, and only then reading the colour of the pixels inside that outline.

A typical pipeline looks like this:

1. Capture. A phone or fixed camera photographs fruit on the tree, in a crate or on a sorting table. A calibration plate (a reference card of known size and colours) is placed in the frame.
2. Detection. An object-detection model (for example a YOLO-family model) finds each apple and draws a box around it.
3. Segmentation. A segmentation model (for example SAM, the Segment Anything Model) traces the exact outline of each fruit, so leaves, branches and background are excluded.
4. Colour correction. The colours of the calibration plate are used to correct for lighting, so a photo taken at 8 am in cloud and one taken at 2 pm in sun can be compared fairly.
5. Measurement. For each fruit, the pixels are converted from RGB into a colour space that matches how we perceive colour — usually HSV (hue, saturation, value) or CIELAB (L*, a*, b*). From there you can calculate:
   • Hue angle of the background colour — moving from green towards yellow as the fruit matures.
   • Blush coverage — the percentage of the fruit's visible surface that falls in the red range.
   • Colour distribution across the sample — not just the average, but how many fruit are behind or ahead.
6. Recording. Results are stored per block, row and date, so trends become visible across the season and between seasons.

This is the same structure Petiole uses for its leaf area pipeline: a calibration plate in the field of view, YOLO detection and SAM segmentation, with each step run as a job in the cloud.

Why not just use the average colour of the image?

Because the average colour of a whole photo mixes fruit, leaves, sky and soil. Even the average of a single fruit can hide what matters: two apples with the same average colour can have very different blush patterns. Segmenting each fruit and measuring the distribution of colour inside it is what makes the numbers meaningful.

What about varieties without a red blush?

For green and yellow varieties — Bramley, Granny Smith, Golden Delicious — there is little or no blush to measure, so the analysis focuses on the background colour and its shift from green to yellow. For bicoloured varieties such as Gala, Braeburn, Cox or Jazz, both background colour and blush coverage are recorded, and they are kept as separate measurements rather than blended into one “redness” score.

How does the starch-iodine test work — and how can AI make it objective?

The starch-iodine test shows how much starch is left in an apple. As apples mature, starch is converted into sugars, and iodine stains starch blue-black, so a less-stained cut surface means a more mature fruit.

The test in brief

1. Pick a representative sample of fruit from the block (different trees, different canopy positions).
2. Cut each apple across its equator (through the “waist”, not top to bottom).
3. Dip or spray the cut face with an iodine-potassium iodide solution and wait a minute or so for the stain to develop. Wear gloves — iodine stains skin and surfaces.
4. Compare each stained face against a reference chart.

Starch usually disappears first around the core and then moves outwards, though the exact pattern varies by variety: some clear in a radial pattern, others in a ring. That is why variety-specific charts exist. Two widely used references are the Cornell Generic Starch-Iodine Index (a 1–8 scale, where 1 is fully stained and 8 is fully clear) and the CTIFL / EUROFRU chart (a 1–10 scale). Harvest targets on these scales depend on the variety and on how long the fruit will be stored, and are usually set with your adviser or packhouse.

Where the traditional test falls short

• Scoring is subjective. Two people looking at the same half can give scores one or two points apart, especially in the middle of the scale.
• Patterns don't always match the chart. Real fruit sits between reference pictures, and scorers round differently.
• Records are thin. Usually only the average score survives in a notebook — the photo, the spread and the odd-looking fruit are lost.

What image analysis adds

Photograph the stained halves on a plain tray with a calibration plate. The model then:

1. Finds and outlines each half.
2. Separates the stained (dark) area from the unstained (pale) flesh.
3. Calculates the stained area as a percentage of the cut face for every fruit.
4. Optionally maps that percentage to the nearest score on your chosen chart.
5. Stores the image, the per-fruit values, the average and the spread together.

The result is a starch reading that is consistent between people and between days, plus a visual record you can go back to when you compare storage results months later.

How do you measure leaf area from a photo?

You measure leaf area from a photo by placing a reference object of known size — a calibration plate — next to the leaf, outlining the leaf with a segmentation model, counting its pixels, and converting pixels into square centimetres using the plate.

The conversion is simple:

Leaf area (cm²) = leaf pixel count × (plate area in cm² ÷ plate pixel count)

Because the plate is in the same photo, at the same distance and angle, this works with an ordinary phone camera and no fixed rig.

Why leaf area matters in apple orchards

• Crop load and fruit quality. Fruit size, sugar and colour all depend on the leaf area available to feed each fruit. Leaf-to-fruit balance is a practical lens on thinning decisions.
• Canopy development. Tracking leaf size through the season shows whether trees are growing as expected after pruning, planting or stress.
• Trials and comparisons. Rootstock, nutrition, irrigation and biostimulant trials need leaf measurements that are faster than a bench leaf-area meter and less subjective than a visual estimate.

Leaf area measurement is the core of what Petiole was built for, and the same calibration-plate approach underpins every other measurement in this article. If you are new to it, start with do I need a calibration plate for Petiole Pro? and our independent validation on cotton leaf area against the LI-COR 3100.

How can AI measure leaf greenness?

AI measures leaf greenness by reading the colour of the segmented leaf, after correcting for lighting with the calibration plate, and expressing it as a greenness index.

Leaf greenness is closely linked to chlorophyll content, and chlorophyll is linked to nitrogen status and overall plant health. The standard field tool is a handheld chlorophyll meter such as a SPAD meter, which clips onto one spot of one leaf at a time. A photo covers the whole leaf — or many leaves — at once.

Common image-based greenness measures include:

• Dark Green Colour Index (DGCI). Calculated from hue, saturation and brightness: DGCI = [(H − 60) ÷ 60 + (1 − S) + (1 − B)] ÷ 3. Higher values mean darker green. It was developed for turfgrass nitrogen status and is widely used for other crops.
• Excess Green (ExG). Calculated from normalised RGB values as 2g − r − b, useful for separating green tissue and comparing greenness.
• CIELAB a*. The a* channel runs from green (negative) to red (positive), so more negative a* means greener tissue.

What leaf greenness can — and can't — tell you

Greenness is an early, cheap signal. Uniform paling of older leaves can point to nitrogen shortage; yellowing between the veins can point to magnesium, iron or manganese issues; and patchy paling can come from root, water or pest stress. Greenness tells you that something is changing and where; it doesn't tell you why. Use it to decide which blocks to sample, then confirm with leaf tissue analysis and your agronomist. For the method in depth, see measuring leaf greenness with the DGCI.

How can AI measure disease pressure on apple leaves?

AI measures disease pressure by outlining each leaf, outlining the lesions or spots on it, and reporting the percentage of leaf area that is diseased. This percentage — disease severity — is the standard way plant pathologists express how badly a leaf is affected.

Diseases where spot area is a useful measure

• Apple scab (Venturia inaequalis) — olive-green to brown, velvety lesions on leaves and fruit; the most important apple disease in the UK and much of Europe.
• Powdery mildew (Podosphaera leucotricha) — whitish, felty coating on leaves and shoots.
• Marssonina blotch (Diplocarpon coronariae) — brown blotches that can lead to early leaf drop; increasingly reported in European orchards.
• Leaf spots from other fungi, and for comparison, physiological spotting and spray damage.

Why area beats a visual score

Disease severity is traditionally estimated by eye, using standard area diagrams or ordinal scales such as the Horsfall–Barratt scale. Visual estimates are quick, but people consistently over- or under-estimate small percentages, and two assessors rarely agree exactly. An image-based measurement gives:

• A continuous percentage, not a class — 3.2% and 4.8% are recorded as different.
• Consistency across assessors, days and seasons.
• Trends — the rate at which diseased area grows after an infection period, which is often more useful than a single snapshot.
• Evidence for spray decisions and trials — comparing blocks, products or timings on the same measured basis.

An important boundary: measuring the area of spots is not the same as diagnosing the disease. Image analysis tells you how much of the leaf is affected. Confirming the cause still needs a trained eye, and sometimes a lab.

We go further into this in AI assessment of necrotic leaf area damage.

How do you take photos that give reliable measurements?

Good measurements start with consistent photos. A simple protocol:

1. Always include the calibration plate, flat and in the same plane as the fruit or leaf.
2. Use soft, even light. Shade or overcast conditions are better than direct sun, which creates glare and hard shadows. Avoid mixed light (half sun, half shade) across the subject.
3. Use a plain background for leaves and cut fruit — a matte board or tray.
4. Keep the camera square on to the subject and at a similar distance each time.
5. Sample representatively. Take fruit and leaves from several trees and from different positions in the canopy (inside and outside, top and bottom, both sides of the row). Most errors come from the sample, not the camera.
6. Label everything. Block, row, variety, date and sample number — in the file name or the app — so results can be trended.
7. Keep the same routine all season, so week-to-week changes reflect the crop, not the method.

The same rules apply to leaves; our guide on how to photograph leaves for accurate leaf area measurement covers them step by step.

Summary: one toolkit, five orchard measurements

Limitations to keep in mind

• Colour is one indicator, not a harvest decision. Combine it with firmness, sugar and starch.
• Lighting and camera differences matter. Without a calibration reference, colour values from different phones or different days are not directly comparable.
• Models need representative examples. A model trained on one variety, one region or one camera may need adjusting for another.
• Area is not diagnosis. Lesion measurement quantifies damage; identifying the cause still requires expertise.
• Sampling still decides accuracy. A perfect measurement of an unrepresentative sample is still unrepresentative.

From LinkedIn