摘要
针对葡萄采摘机器人对绿色葡萄果实和被树枝或树叶遮挡的葡萄果实的检测准确率不高问题,搭建了基于改进YOLOv5s(YOLOv5s-MM)的葡萄采摘机器人果实检测新模型。该模型首先在YOLOv5s原有模型中添加垂直和水平旋转操作,以有效提高模型的泛化能力。然后在YOLOv5s原有模型的Feature Map中增加偏心纠正模块和小目标检测层,提高模型的检测精度和速度。最后利用包含绿色葡萄和被遮挡葡萄的图像数据库对所构建的模型进行了训练和测试,验证了模型的有效性。对测试集中葡萄图像的测试结果表明,所搭建的葡萄采摘机器人果实检测新模型能够比YOVOv5、YOLOv4-Grape和基于边缘图像处理和几何形态学的方法更加准确地检测绿色葡萄果实和被树叶、树枝等遮挡的葡萄果实。所搭建模型可为葡萄采摘机器人实时准确检测绿色葡萄果实和被树枝或树叶遮挡的葡萄果实提供技术支持。
To solve the problem that the detection accuracies of green grape fruits and grape fruits obscured by branches or leaves of grape picking robot is not high,a novel fruit detection model for grape picking robot based on an improved YOLOv5s(YO-LOV5S-MM)is constructed in this paper.Firstly,vertical and horizontal rotation operations are applied for the original YO-LOv5s model to improve the generalization ability of the model.Then,an eccentric correction module and a small target detection layer are added to the Feature Map of the original YOLOv5s model to improve the detection accuracy and speed of the model.Fi-nally,based on the train and test experiments on an image databases containing green grape fruits and occluded grape fruits,the effectiveness of the model is verified.The test results of grape images in the test set show that the novel grape fruit detection model constructed in this paper can more accurately detect green grape fruits and grape fruits that are occluded by leaves or branches than YOLOv5s,YOLOv4-Grape and method based on edge image processing and geometric morphology.The constructed model can provide technical support for grape picking robot to accurately detect green grape fruits and grape fruits that are obscured by branches or leaves in real time.
作者
李晋超
朱晓军
LI Jinchao;ZHU Xiaojun(Shanxi Electromechanical Vocational and Technical College,Shanxi Changzhi 046011,China;Taiyuan University of Technology,Shanxi Taiyuan 030024,China)
出处
《机械设计与制造》
北大核心
2025年第11期359-364,共6页
Machinery Design & Manufacture
基金
山西省教育科学“十三五”规划课题(GH-19225)
山西省回国留学人员科研资助项目(2020-154)。
