基于YOLOv11-FS模型的柑橘花粉活力检测OA

A dataset of detecting pollen viability in citrus was constructed to promote the breeding of citrus seedless varieties and meet the demand for high-quality production of citrus.A new model of detecting pollen vi-tality was proposed by improving the YOLOv11 deep neural network.A dataset suitable for detecting pollen vi-tality was established through manual collection and expert annotation.The YOLOv11 deep neural network was improved to develop the YOLOv11-FS model.The Focal-EIOU loss function was used to replace the EIOU loss of YOLOv11 to enhance detection performance on imbalanced samples during the process of detection.Soft-NMS was incorporated to improve the precision of detection boxes,overcoming challenges including obvi-ous clustering,small size,and complex backgrounds of pollen grains in citrus.Results showed that the improved YOLOv11-FS model performed excellently in the task of detecting pollen,the predicted pollen viability deviates from the true pollen viability by only 0.70 percentage points.The recall,precision,and F1-score of detecting fer-tile pollen and sterile pollen reached 98.76%,99.67%,99.22%,and 94.87%,98.89%,96.84%,respective-ly,meeting the basic needs of detecting pollen vitality.It is indicated that the method of detection establshed can achieve efficient and accurate identification of pollen vitality in citrus.