Quantifying wheat spikes through smartphone camera and YOLOv5 under open field conditions

A wheat spikes quantification model was developed and tested under open field measurement conditions using a smartphone camera. YOLOv5 was trained using wheat spike images obtained from the Global Wheat Head Dataset. The model was tested through a total of 700 images captured from the study field at different times of day, phone measurement angles and wheat spike maturity stages. The most accurate estimations were obtained at 6 pm and phone angle of 15-20°. Such results prove the robustness and transferability of the YOLO-based model in wheat spike detection which can help farmers and plant breeders in field measurements.