自然复杂环境下油茶果识别的重参数化算法OA
RepVGG algorithm for Camellia oleifera fruits recognition in natural complex environment
针对自然环境下油茶果机器采摘识别任务中存在果实密集粘连、枝叶遮挡、果实颜色差异及光照不均等挑战,并结合当前相关研究在复杂场景下检测精度与鲁棒性不足的问题,提出基于 YOLOv8n改进的 YOLOv8—COD模型.该模型对 C2f模块中的超参数进行调整,同时融入轻量化卷积模块,使用重参数化模块代替主干网络中的卷积模块,在提升模型检测精度的同时保持计算效率;在特征融合模块中添加 GAM(Global Attention Mechanism)注意力机制,并使用GIoU—Focal 替换 CIoU,有助于模型聚焦油茶果,提高模型在果实遮挡、粘连等情况下的识别率.相比于传统YOLOv8n,该模型的精确率 P、召回率 R、平均精度均值 mAP 分别提升 0.2%、3.3%、2.1%.在复杂自然环境中,YOLOv8—COD的漏检概率相比于 YOLOv8n有明显下降,检测精度有所提升,能够有效地实现油茶果的检测识别.
To address the challenges in the machine picking recognition task of Camellia oleifera fruits in natural environments,such as dense fruit adhesion,leaf and branch occlusion,fruit color difference,and uneven lighting,and in light of the current research issues of insufficient detection accuracy and robustness in complex scenarios,an improved YOLOv8n model,namely YOLOv8—COD,is proposed.In this model,hyper parameters in C2f module are adjusted and lightweight convolutional module is integrated.Heavy parameterization module is used to replace convolutional module in backbone network,so as to maintain computational efficiency while improving model detection accuracy.Adding Global Attention Mechanism(GAM)into the feature fusion module and replacing CIoU with GIoU—Focal can help the model focus on camellia fruit and improve the recognition rate of the model under the conditions of fruit occlusion and adhesion.Compared with the traditional YOLOv8n,its precision rate,recall rate,and mAP are increased by 0.2%,3.3%,and 2.1%respectively.In complex natural environment,the missed detection probability of YOLOv8—COD decreased significantly compared with YOLOv8n,and the detection accuracy was improved,which can effectively realize the detection and identification of Camellia oleifera fruits.
Xiao Shenping;Deng Hongjin;Zhao Qianying;Chen Yongzhong
School of Electrical and Information Engineering,Hunan University of Technology,Zhuzhou,412007,China||Key Laboratory for Electric Drive Control and Intelligent Equipment of Hunan Province,Zhuzhou,412007,ChinaKey Laboratory for Electric Drive Control and Intelligent Equipment of Hunan Province,Zhuzhou,412007,China||School of Railway Transportation,Hunan University of Technology,Zhuzhou,412007,ChinaKey Laboratory for Electric Drive Control and Intelligent Equipment of Hunan Province,Zhuzhou,412007,China||School of Railway Transportation,Hunan University of Technology,Zhuzhou,412007,ChinaNational Engineering Research Center for Oil Tea Camellia,Hunan Academy of Forestry,Changsha,410100,China
农业科技
油茶果YOLOv8n检测识别YOLOv8—COD重参数化
Camellia oleifera fruitsYOLOv8ndetection and recognitionYOLOv8—CODRepVGG
《中国农机化学报》 2026 (2)
78-85,8
国家重点研发计划项目(2019YFE0122600)2024 年度十大技术攻关项目(2024NK2020)
评论