Application of YOLO-based deep learning and UAV imagery in hybrid maize seed production

ABSTRACT In maize breeding programs, emasculation is a critical step in the production of hybrid seeds, occurring prior to anthesis. Therefore, an accurate identification of tassels is paramount for the efficient execution of detasseling. However, tassel detection poses significant challenges due to environmental variability in production fields, the spatial resolution of aerial images, lighting conditions, and the phenotypic variability of the plants. Traditional methods are also predisposed to subjectivity and detasseling errors. This study proposes a solution to these challenges by integrating a remote sensing platform, utilizing an unmanned aerial vehicle (UAV), with a deep learning (DL) algorithm to enhance the efficiency of maize tassel detection. The UAV captured the red-green-blue (RGB) images spectral region at an altitude of 20 m, and the YOLOv4 algorithm was employed to develop the tassel detection model. The image database was constructed using preprocessing strategies to augment the dataset. The detection model based on the YOLOv4 architecture demonstrated satisfactory results, with an overall accuracy of 0.89, ranging from 0.84 to 0.92, depending on the levels of detasseling errors and the morphological characteristics of maize tassels. This study underscores the applicability of YOLOv4 in detecting plants with detasseling errors in seed production fields, thereby contributing to the purity and quality of hybrid seeds.