Can a phone measure fruit as well as a caliper?
Phones are now accurate to a few millimetres on visible fruit. The details that decide whether that is good enough are tilt, occlusion and calibration.
How accurate is it?
A review of orchard sizing found that an error below 5 mm is now routinely achieved for fruit that are not hidden by leaves.1 A smartphone application tested in the laboratory gave diameter errors (RMSE) of 3.8 mm for mandarin and 2.4 mm for navel orange, and 2.0 to 2.1 mm across two different phones against a caliper.2 A Mask R-CNN method on iPhone 12 Pro images of satsuma, using a small reference plate, reached a mean absolute error below 1 mm.3
The caliper is not perfect either
A caliper can be read to below a millimetre, but where it is placed on the fruit adds a larger sampling error.1 The equator of a mandarin is not a perfect circle, so even manual measurement takes skill.3
What goes wrong
- Tilt. In one study, a camera tilt above 14° raised the distance error to 12 mm, which means about 6 percent error in fruit size.2
- Occlusion. Partly hidden fruit should be counted, but for sizing they should be rejected or reconstructed from the visible part.1
- Depth alone. In tests on cylindrical targets, iPhone 13 Pro LiDAR scans showed diameter variability of up to 19.6 percent, so raw depth needs to be combined with image segmentation.4
- Class boundaries. A citrus model with a 2.15 mm mean error still graded only 57 percent of fruit into the correct size class, because small errors matter at class edges.5
What this means in the orchard
Hold the phone square to the fruit, at a controlled distance, and measure only fruit that are fully visible. Calibrate against calipers on a handful of fruit per block. Report size distributions with confidence intervals rather than single values.
Sources
- Neupane C., Pereira M., Koirala A., Walsh K.B. (2023). Fruit sizing in orchard: a review from caliper to machine vision with deep learning. Sensors 23(8):3868. doi
- Wang Z., Koirala A., Walsh K., Anderson N., Verma B. (2018). In field fruit sizing using a smart phone application. Sensors 18(10):3331. doi
- Komiya et al. (2025). Development of a fruit size estimation method using Mask RCNN for water stress estimation in Satsuma mandarin. PLoS ONE 20(7). doi
- Sensors 25(18):5629 (2025). Photogrammetric and LiDAR scanning with iPhone 13 Pro: accuracy, precision and field application on hazelnut trees. doi
- Liu H., Cao Z. et al. (2026). PMDS-YOLO: a lightweight deep learning model for non-destructive citrus fruit diameter estimation and quality grading. Foods. PMC