基于YOLOv11-FCBG的水稻种子发芽状态检测研究OA
水稻种子发芽率是评估种子品质和活力的重要指标,决定育秧质量和最终产量.传统人工目视检测有效率低、标准不统一等缺点.针对培养皿中水稻种子发芽检测的场景,为解决原始YOLOv11 模型在小目标检测中由于细微特征提取不足,特征融合不充分导致的漏检、误检的问题,该文以YOLOv11n为基础,引入FasterNet、C2DA、GLSA、BiFPN模块,建立YOLOv11-FCBG模型,并采用Roboflow开源水稻种子数据集进行训练.结果显示,YOLOv11-FCBG模型的平均精确率、召回率、mAP50分别为 0.844、0.88、0.921,比原版YOLOv11 模型,精确率提升 3.05%、mAP50 提升 1.77%,可提升水稻种子发芽检测能力.
Rice seed germination rate is an important indicator to evaluate seed quality and vitality,and determines seedling quality and final yield.Traditional manual visual inspection has shortcomings such as low efficiency and inconsistent standards.Aiming at the scene of rice seed germination detection in a petri dish,in order to solve the problem of missed detection and false detection caused by insufficient extraction of fine features and insufficient feature fusion in small target detection,this paper is based on YOLOv11n and introduces FasterNet,C2DA,GLSA,and BiFPN modules to establish a YOLOv11-FCBG model,and uses the Roboflow open source rice seed dataset for training.The results showed that the average accuracy,recall,and mAP50 of the YOLOv11-FCBG model were 0.844,0.88,and 0.921 respectively.Compared with the original YOLOv11 model,the accuracy was increased by 3.05%and mAP50 was increased by 1.77%,which improved the ability to detect rice seed germination.
梁澄河;李佳宝
广西职业技术大学,南宁 530226广西职业技术大学,南宁 530226
农业科技
水稻种子发芽检测YOLOv11-FCBG小目标检测特征融合
rice seedsgermination detectionYOLOv11-FCBGsmall target detectionfeature fusion
《智慧农业导刊》 2026 (14)
13-17,5
2025年度广西高校中青年教师科研基础能力提升项目(桂教科研[2025]1 号,2025KY1409)广西职业技术学院校级课题(桂职院[2023]121 号,231205)
评论