基于改进YOLOv8-Pose的码垛快速识别与抓取点检测  

Fast Palletizing Recognition and Grab Point Detection Based on Improved YOLOv8-Pose

作　　者：郭忠峰[1] 王健鹏 杨钧麟 杨春源 GUO Zhongfeng;WANG Jianpeng;YANG Junlin;YANG Chunyuan(School of Mechanical Engineering,Shenyang University of Technology,Shenyang 110870,China)

机构地区：[1]沈阳工业大学机械工程学院,沈阳110870

出　　处：《组合机床与自动化加工技术》2024年第11期125-129,共5页Modular Machine Tool & Automatic Manufacturing Technique

摘　　要：针对码垛场景中在仓库内对米袋和面袋的识别与抓取点检测的任务,提出了一种基于改进的YOLOv8-Pose的轻量化快速检测算法模型。其基于YOLOv8-Pose,使用若干个ShuffleNetv2模块取代原Darknet主干网络,降低模型大小;添加SimAM注意力机制,提升目标特征提取能力。通过对比实验表明,该模型在不牺牲准确性的前提下可提升模型的识别速度。模型在自制数据集中的平均精度达到了93.7%,检测速度达到了62 fps,优于常见模型。证明该模型能够实现复杂场景下的抓取点识别,且该轻量化模型能够适用于嵌入式硬件,降低设备成本。Aiming at the task of recognition and grab point detection of rice bags and face bags in warehouse in palletizing scene,a lightweight fast detection algorithm model based on improved YOLOv8-Pose is proposed.Based on YOLOv8-Pose,this algorithm uses several ShuffleNetv2 modules to replace the original Darknet backbone network and reduce the model size.SimAM attention mechanism is added to improve the ability of target feature extraction.The comparison experiment shows that the model can improve the recognition speed without sacrificing the accuracy.The lightweight model can be applied to embedded hardware.The average accuracy of the model in the self-made data set reaches 93.7%,and the detection speed reaches 62 fps,which is better than the common model.The experimental results show that the model can realize the recognition of grab points in complex scenes.The lightweight model can be applied to embedded hardware and reduce equipment cost.

关 键 词：抓取点检测 YOLOv8-Pose ShuffleNetv2 轻量化网络结构 

分 类 号：TH165[机械工程—机械制造及自动化] TG659[金属学及工艺—金属切削加工及机床]