Maize Ear Sensing for On-Farm Yield Predictions
Nondestructive depth sensing that turns each captured husk-on maize ear into an instant, per-plant grain-yield estimate.
Abstract
We introduce the first fully on-field pipeline that estimates maize-ear length, width and volume from a single RGB + depth capture and immediately forecasts grain yield per plant. A YOLOv12n-seg model isolates the ear in unconstrained lighting, a bespoke network (EVNet) regresses volume from the segmented point cloud, and gradient-boosted trees convert morphology into yield.
On Kansas field data we reach 98.6% mAP@0.5 for segmentation, 28.9 ml RMSE for volume, and 13.9 g RMSE for yield (ideal) / 24.1 g (real). The pipeline runs in ≈1 s per image, needs no destructive sampling, and the images, code, and trained weights are open-sourced.
Real-time field deployment
Maize ear detection & yield prediction
Our Pipeline
Key Results
Segmentation
95.8% precision/recall
mAP@0.5 = 98.6%
1.11s per image
Volume (Real-world)
R² = 0.88
RMSE = 28.9 ml
Yield (Ideal)
R² = 0.96
RMSE = 13.9g
Yield (Real-world)
R² = 0.89
RMSE = 24.1g
Why it Matters
First non-destructive ear yield predictor deployable in the field.
Open dataset & code (CornDepth) to accelerate follow-up work.
Bridges phenotyping & on-farm decision-making for breeders and agronomists.
Resources
Citation
@InProceedings{Cisdeli_2025_CVPR,
author = {Cisdeli, Pedro and Santiago, Gustavo Nocera and Mandrini, German and Ciampitti, Ignacio},
title = {Maize ear sensing for on-farm yield predictions},
booktitle = {Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR) Workshops},
month = {June},
year = {2025},
pages = {5402-5411}
}