Apple Detection and Color Analysis Checklist
What this check is for
This checklist counts the apples in a photo and sorts each one by skin color: red, green, yellow or bicolor. Use it for quick orchard samples, spot checks on a packing line, a store display count or a school project. You can also use it to test how well an AI image model counts and sorts fruit. A photo check gives only a rough result. It shows only one side of each apple, and the result depends a lot on the light. Use the result as an estimate, not as a grade.
Take photos that can be checked
- Use daylight or neutral white light. Warm indoor bulbs can make green apples look yellow, and colored light changes all the colors.
- Use a plain white or gray surface. Do not use red or green tablecloths and crates.
- Put a white sheet of paper or a gray card in the frame. It shows you if the photo has a color cast (a tint over the whole photo).
- Put the apples in a single layer with little overlap. Take the photo from above. For a full bin, take several photos of the top layer. Write down that the count covers only the visible apples.
- Keep the whole group in focus. Do not let strong shadows hide apples at the edges.
Set your color rules first
Write down the color rules before you count. Then two people (or a person and an AI) put the same apple in the same class. For example:
| Class | Rule (example, change it to fit your needs) |
|---|---|
| Red | Red covers more than about three quarters of the visible surface |
| Bicolor | Red covers part of the surface over a green or yellow base |
| Green | Little or no red, with a green base color |
| Yellow | Little or no red, with a yellow base color |
| Cannot tell | Too dark, blurred, or mostly hidden |
Red coverage is the share of the visible surface that is red. If you also want an estimate of red coverage, use bands. Example bands are 0 to 25, 25 to 50, 50 to 75 and 75 to 100 percent. It is easier for people to agree on a band than on an exact percentage.
The checklist
- Count every apple. Include apples that the frame cuts off at the edges and apples that other apples partly hide. Where photos overlap, do not count the same apple twice.
- Make sure that each one is an apple. Look at the shape and the stem cavity. Pears, peaches, tomatoes and decorative fruit often get into the photo too.
- Give each apple a number. Number them left to right, top to bottom. Then anyone can find “apple 7” and look at your decision.
- Give each apple a color class with your color rules.
- If you need it, estimate the red coverage band.
- Look at the white or gray reference. If it looks yellow or blue, write a note. The light can shift the color classes.
- Write down visible defects, such as bruises, cuts, rot or spots, in a separate column. Do not let them change the color class.
- Mark each unclear apple as “Cannot tell.” Do not guess.
If you grade apples for sale in the US, the formal color rules are stricter. For 12 listed varieties, the U.S. Standards for Grades of Apples set a minimum share of red. This share is the part of the surface that must show a good shade of red. For Red Delicious, the minimum is 66 percent for U.S. Extra Fancy, 40 percent for U.S. Fancy and 25 percent for U.S. No. 1. Varieties that the standard does not list have no color requirement for these grades (USDA AMS standards, section 51.305). A photo of one side cannot show the whole surface. Use photo counts to sort and sample. Grade the fruit in your hand.
Record the result
Make a count sheet for each photo:
| # | Position | Color class | Red coverage (visible) | Defects | Note |
|---|---|---|---|---|---|
| 1 | Top row, left | Red | 75 to 100% | None | |
| 2 | Top row, second | Bicolor | 25 to 50% | Small bruise |
Then make a summary:
| Color class | Count | Share of total |
|---|---|---|
| Red | ||
| Bicolor | ||
| Green | ||
| Yellow | ||
| Cannot tell | ||
| Total | 100% |
Write the photo date and the light conditions next to the summary. You can compare counts from different days only when you take the photos the same way.
Use this checklist with Claude or ChatGPT
Image models in Claude and ChatGPT can count objects and describe their colors. This makes them useful for a first pass. Phone photos can contain location data. If you want to keep the place private, remove the location data before you upload the photo. Read the data settings of your workspace to learn how the service stores and uses uploaded images.
Attach the photo or photos and paste this prompt:
Count the apples in the attached photo and classify each one by color.
Color rules:
[paste your color rules and coverage bands]
1. Number every apple left to right, top to bottom, and describe
where each one is so I can find it.
2. Include apples that are cut off at the edges or partly hidden.
Say which ones are partly hidden.
3. For each apple, give the color class, the red coverage band of the
visible surface, any visible defects, and one sentence explaining
your call.
4. Say whether the light or background seems to shift the colors.
5. Give a summary table: count and share for each color class.
Mark any apple you are not sure about as Cannot tell. At the end, list
what you could not determine, such as hidden or blurred fruit.Image models can miscount crowded or overlapping fruit. They can also merge two apples into one or count a shadow as an apple. Count the apples in at least one photo yourself. Compare your count with the model count. Look at every apple marked Cannot tell. If you use the numbers to buy, sell or grade fruit, a person must make sure that they are correct.
A docs portal with Ask AI
and support tickets
Connect a repository, or describe your product to start. Each docs site has search, Ask AI, an API console, an MCP server and support tickets.
- Publishes from GitHub on every push
- Works with Mintlify repositories
- API console from your OpenAPI file
- MCP server and llms.txt for AI agents
- Ask AI and support tickets
- Analytics for people and AI agents