Measuring Popcorn Traits with Computer Vision
How breeders can turn flake shape, expansion, and unpopped kernels into objective, comparable data.
Many of the traits that determine popcorn quality only become visible after popping. Two breeding lines may look similar as kernels but produce very different results when heated: one may expand consistently into large flakes, while another produces smaller flakes, mixed shapes, or a high number of unpopped kernels.
This creates a practical challenge for breeding programs. Important post-pop traits must be evaluated across many lines, replicates, seasons, and environments. When that evaluation depends mainly on visual scores, results can vary between people, and valuable differences within a sample can be missed.
Computer vision provides a more objective way to evaluate the complete popped sample. Instead of assigning one general score, it measures individual flakes and kernels and records the distribution of traits across the sample.
The traits breeders need to see after popping
Flake shape
Popcorn flakes are commonly grouped into two broad commercial types:
- Ball or mushroom flakes have a compact, rounded structure. They are generally more resistant to handling and are often preferred for coated popcorn products.
- Butterfly flakes expand into irregular shapes with one or more wings. They create the open appearance associated with ready-to-eat and cinema-style popcorn, but they can be more vulnerable to breakage.
These categories are useful, but a breeding sample rarely produces only one shape. It produces a distribution. Even a line described as butterfly popcorn may contain flakes with different numbers of wings, different levels of symmetry, and a wide range of sizes.
That distribution matters. Research has shown that flake morphology is related to expansion volume and packing behaviour. A single average or visual score cannot fully describe those differences.
Expansion and unpopped kernels
Expansion volume is one of the main traits used to evaluate popcorn performance. The number of unpopped kernels is equally important because it represents product that did not convert into saleable volume.
The two traits should be considered together. A sample with high average expansion but a large unpopped fraction may be less attractive than a sample with slightly lower expansion and more consistent popping.
Both are also affected by the test protocol. Kernel moisture, sample mass, popping method, and kernel size can influence the result. For breeding comparisons to be meaningful, these conditions must be standardised.
Pericarp presence
The pericarp helps the kernel retain pressure during heating. Damage to it can allow steam to escape too early, resulting in a small flake or an unpopped kernel.
After popping, fragments of pericarp may remain on the flakes. This is often assessed using a visual rating scale, but those ratings can vary among evaluators. Image analysis can instead separate brighter endosperm areas from darker pericarp and hull areas and consistently report their relative proportions.
This is a visual measurement of pericarp presence. It does not measure tenderness, flavour, or overall eating quality, which still require sensory or laboratory evaluation.
From one score to a complete sample profile
Computer vision changes the unit of analysis. The breeder is no longer limited to one score for the sample. Each visible object can be detected, classified, and measured.
Get Yaniv Noema’s stories in your inbox
Join Medium for free to get updates from this writer.
For a whole popped sample, an imaging system can:
- Separate popped flakes from unpopped kernels
- Count each class
- Measure flake area, length, width, and shape
- Classify ball and butterfly forms using defined criteria
- Quantify bright and dark surface areas
- Save the image together with the results
The output is not simply a pass or fail decision. It is a measurable profile of the sample.
Why distributions are more useful for selection
Two breeding lines can have the same average flake size while behaving very differently. One may produce flakes within a narrow, consistent range. The other may combine very large flakes with many small ones.
Those samples are not equivalent for processing or packaging, even if their averages are identical.
Object-level analysis allows breeders to compare:
- The percentage of each flake shape
- The spread of flake sizes
- The proportion of unpopped kernels
- Consistency across replicates
- Stability across seasons or locations
This makes selection more application-specific. A program targeting coated popcorn can prioritise a high and stable proportion of compact flakes. A program targeting cinema-style popcorn can focus on butterfly morphology, flake size, and expansion. In both cases, consistency can be evaluated alongside the average result.
More consistent data across a breeding program
The main value of computer vision is not that it replaces every popcorn test. It is that it makes visible traits easier to measure repeatedly and at scale.
Using fixed imaging conditions and defined classification rules helps reduce differences between operators. Saving the sample image also creates a traceable record. If a classification rule changes later, archived images can be reviewed or analysed again without repeating the popping test.
This is especially useful when breeding decisions depend on comparisons across large populations, multiple environments, or several years of trials.
Where Vibe fits
Vibe Imaging Analytics develops whole-sample imaging systems for grain and seed analysis. The QM3i can analyse popped and unpopped popcorn samples, classify flake shapes, measure individual objects, quantify bright and dark areas, and retain the image and results for later review.
For breeding programs, this turns post-pop visual evaluation into structured phenotype data that can be compared across samples, replicates, and seasons.
See how Vibe applies imaging analytics to popcorn breeding.
About the author
The author works at Vibe Imaging Analytics, which develops the imaging system referenced in this article.
Research note: Discussion of flake morphology, expansion, packing behaviour, and sensory response is based on popcorn research published in Cereal Chemistry and associated University of Nebraska work. The relationship between pericarp integrity, pressure, and popping performance is supported by published popcorn physiology research.
