Detecting tassels in UAV imagery with Tassel-YOLOv12n model: A case study of adapted maize varieties in the Huaibei Plain, China

The appearance of maize tassels marks the onset of the tasseling stage, which serves as an important indicator for crop management, variety selection and yield forecasting. The tassel is located at the top of the plants, making them suitable for large-scale remote sensing by UAV platforms. The tassel morphology varies depending on factors including cultivar, irrigation, fertilization and environments. In the field, tassel detection using UAV imagery remains challenging due to some factors, such as the small size of ears, frequent occlusions and low color contrast. To address these challenges, Tassel-YOLOv12n model was developed in this study based on maize_hfnu dataset, which was established with the UAV imagery of typical adapted maize varieties in the Huaibei plain, China. The proposed model employed a combined structure of feature pyramid network and path aggregation network to enhance tassel detection capability. The rfficient multi-scale attention mechanism module and the small object detection layer were also introduced to extract the features of tassels in field environments. During experimental evaluation, our proposed Tassel-YOLOv12n algorithm achieved the [email protected] value of 87.7 and the [email protected] value of 40.4, which is better than well-known object detection approaches. The result showed that Tassel-YOLOv12n achieves strong performance in both counting accuracy and inference speed.