CornPheno provides corn breeders with a method to measure key ear traits — such as kernels per ear, rows per ear, and kernels per row — directly in the field. Using AI-based image analysis, the smartphone platform aims to address constraints of traditional phenotyping approaches, which are often labour-intensive, prone to manual counting errors, and dependent on costly equipment or controlled imaging conditions. It allows ear traits to be recorded and analysed in outdoor environments where lighting and backgrounds vary.
Corn is a major global crop used for food, feed, and industrial purposes. Phenotyping traits such as kernels per ear and rows per ear is important for selecting high-performing varieties, but conventional methods can be slow and resource-intensive. Although AI and computer vision have improved automation, many existing tools remain costly or perform inconsistently under field conditions. A press release stated that CornPheno is designed to collect trait data under real-world conditions using a smartphone.
A study published in Plant Phenomics on 15 October 2025 by Hao Lu’s team at Huazhong University of Science and Technology reported that CornPheno can measure key corn ear traits using smartphone images, supporting field-based assessments and breeding decisions.
To evaluate performance, the researchers compared CornPheno with five plant counting models — BCNet, CSRNet, P2PNet, TasselNetV2, and CCTrans — using Mean Absolute Error (MAE), Mean Squared Error (MSE), and the coefficient of determination (R²). CornPheno combines two models: CornPET for kernel detection and Unicorn for row detection. CornPET uses crowd-localisation techniques to detect kernels in dense ear images, including in field conditions, while Unicorn addresses issues such as row distortion, missing kernels, and misalignment.
The system was tested in machine, indoor, and field settings. The researchers reported an R² of 0.7641 for kernels per ear, higher than the next-best model, P2PNet (R² = 0.6884). In field conditions, CornPheno achieved an R² of 0.8190, compared with 0.8322 under controlled conditions. For rows per ear, it correctly detected rows 80.3% of the time with an MAE of 0.234. For kernels per row, it achieved an F1-score of 0.9418, with precision of 0.9796 and recall of 0.9091.
CornPheno is integrated into OpenPheno, a WeChat-based mini-program used for in-field phenotyping on smartphones. The researchers also developed the CKC-Wild dataset, consisting of 1,727 annotated corn ear images captured across varied outdoor environments. Reported results on this dataset indicate the system can count kernels and detect rows under different field conditions, supporting its use in larger-scale breeding and research applications.
