How does a plant identifier decide what plant it is?
A plant identifier decides which plant is in a photo. It compares the leaves, flowers and bark with patterns it learned from many labeled plant photos, and gives its best guess a confidence score. If the score is over a line, it shows the name. A high score is not a promise: it can still be wrong.
In one line: A program that looks at a photo of a plant, guesses its name and gives the guess a confidence score.
Last reviewed: 25 October 2026 3 min read
What decision does it make?
The identifier makes two choices. First it picks the plant name that fits the photo best. Then it decides if it is sure enough to show that name, or should say it is not sure.
| Decision | Possible answers |
|---|---|
| What plant is this? | Its best guess from the plants it knows, with a confidence score. Over the line: show the name. Under the line: say it is not sure. |
What does it look at?
- Leaves. The shape and edges of the leaves. In one study of trees, leaf photos gave better results than bark photos.
- Flowers and fruit. Photos that show flowers or fruit gave better results in one study.
- Bark. Bark gives clues too, but bark photos alone were found to be much less accurate than leaf photos.
- Photo quality. A photo that is in focus and well lit gave better results, so a blurry or dark photo makes guesses weaker.
How does it learn the rule?
- Collect labeled photos. People gather many plant photos and label each one with the plant's right name.
- Learn from the photos. The program learns from photos that already have plant names.
- Score the guess. For a new photo, it picks the plant name that fits best and gives that guess a score.
- Set the line. People choose how high the score must be before the name is shown, and keep checking the mistakes.
Where can it go wrong?
- A wrong name shown. A mushroom is not a plant, but a plant identifier may still guess a plant name for it.
- A right name held back. A correct guess about a plant in a dark room can score under the line, so it says it is not sure.
- Look-alikes. Some plants have very similar leaves, so one plant can be mistaken for another. Never eat or touch a plant just because a program named it.
- No line is perfect. Move the line up and fewer wrong names are shown, but more right names are held back.
Try it
Toy example: move the line to choose how sure the identifier must be before it shows a name. Can you find a line with no mistakes?
Toy example with invented scores: ten plant photos described in words, the confidence score of the best guess, and whether that guess was right.
| Item | Values | Right bin |
|---|---|---|
| Sunflower in full sun, whole flower | Confidence score: 95 | Shows the name |
| Oak leaf laid flat on paper | Confidence score: 90 | Shows the name |
| Daisy, close up | Confidence score: 85 | Shows the name |
| Tomato plant with red fruit | Confidence score: 80 | Shows the name |
| Mushroom on a log (not a plant) | Confidence score: 70 | Says it is not sure |
| Rose leaves, no flower | Confidence score: 60 | Shows the name |
| Blurry photo of a fern | Confidence score: 55 | Says it is not sure |
| Seedling with two tiny leaves | Confidence score: 45 | Says it is not sure |
| Houseplant in a dark room | Confidence score: 40 | Shows the name |
| Weed half hidden by grass | Confidence score: 25 | Says it is not sure |
Check it
What does the confidence score tell you?
Answer: How sure the identifier is of its guess. It is how strongly the photo matches what it learned. It is not a promise.
You move the line higher. What happens?
Answer: Fewer wrong names are shown, but more right names are held back. A higher line is more careful, so it also says "not sure" more often when it was right.
An identifier names a plant with a score of 95. Is it safe to eat?
Answer: No, never eat a plant because a program named it. High scores can still be wrong, and plants can look alike. Ask a grown-up who knows plants.
For older readers
A plant identifier is a classifier trained with supervised learning on labeled photos: it gives its best guess a score. The confidence threshold turns that into "show" or "not sure". Raising it lowers false positives (wrong names shown) and raises false negatives (right names held back). A photo of something it was not trained on, like a mushroom, may still get a plant name.
Questions
Take a clear, bright photo of just the plant, close enough to see the leaves, and the flower if it has one. Leave people out of the picture.
Its best guess scored under the line, maybe because the photo was blurry or the plant looked like several others.
No. Never eat or touch a plant just because a program named it. Ask a grown-up who knows plants.
Sources
- University of Washington CSE 416: Classification (confusion matrix and changing the threshold) (opens another website)
- University of Massachusetts Amherst: Introduction to Supervised Learning (opens another website)
- University extension: how accurate are plant identification apps? (opens another website)
- University study: plant identification apps and photo quality (opens another website)