摘要
针对复杂场景中机器人的无序抓取需要,提出一种两阶段的抓取检测算法。改进YOLOv5的网络模型,在多尺度特征融合上将浅层位置信息和深层语义信息进行注意力融合,提高多尺度目标的检测能力;将排斥因子引入损失函数中,提高了模型在遮挡环境下的鲁棒性;在目标检测后对抓取目标边界框进行裁切处理,避免了抓取检测过程中其余目标的干扰;改进抓取检测算法,引入CSP结构和注意力机制,提高了模型的特征提取能力。在真实环境下针对随意摆放的多目标遮挡物体进行抓取实验,结果表明:机器人抓取成功率为95%。
A two-stage grasp detection algorithm is proposed for the disorderly grasping needs of robots in complex scenes.The network model of YOLOv5 is improved by attention fusion of shallow location information and deep semantic information on multi-scale feature fusion to improve the detection of multi-scale targets.The rejection factor is introduced into the loss function to enhance the robustness of the model in occlusion environment.The grasp target bounding box is cropped after the target detection to avoid the interference from the rest of the targets during the grasp detection process.The grasp detection algorithm is improved by introducing the CSP structure and attention mechanism to improve the feature extraction ability of the model.In grasping multi-target obscured objects randomly placed in a real environment,the results show that the robot has a 95%success rate.
作者
朱文磊
董淑宏
张洪
于培师
徐稳
ZHU Wenlei;DONG Shuhong;ZHANG Hong;YU Peishi;XU Wen(School of Mechanical Engineering,Jiangnan University,Wuxi 214122,China;Jiangsu Key Laboratory of Advanced Food Manufacturing Equipment&Technology,Jiangnan University,Wuxi 214122,China)
出处
《机械制造与自动化》
2024年第5期218-223,共6页
Machine Building & Automation
基金
国家自然科学基金项目(11972171)。
关键词
调压阀
目标检测算法
轻量化
重参数化
特征融合
pressure regulating valve
target detection algorithm
lightweight
re-parameterization
feature fusion
