摘要
针对在移动设备和嵌入式设备等资源受限的环境中,机器人不易实时准确抓取物体的问题,提出一种基于深度可分离卷积残差模块的卷积神经网络(CNN)模型。该模型充分利用相机颜色和深度信息,以RGB-D图像作为网络输入,直接对逐个像素点完成抓取预测。利用深度可分离卷积替代传统残差结构中的标准卷积层,构建出深度可分离卷积残差模块,在不降低网络性能的基础上减少模型参数,网络模型大小仅为2.3 MB。最后,在Cornell抓取数据集上进行实验,准确率达到97.7%,检测速度为58 fps。
Aiming at the problem that it is difficult for robots to real-time and accurately grasp object in resource-limited environments such as mobile and embedded devices,a convolutional neural network(CNN)model based on depthwise separable convolution residual block is proposed This model fully utilizes camera color and depth information,using RGB-D images as network inputs,to directly achieve grasping predict for each pixel point.DSC is used to replace the standard convolution layer in the traditional residual structure,construct a DSC residual block,reduce the model parameters without decreasing the network performance.The size of the network model is only 2.3 MB.Finally,experiments are conducted on the Cornell Grasping Dataset,an accuracy of 97.7%and a detection speed of 58 fps are achieved.
作者
平路静
马行
穆春阳
姜谱照
PING Lujing;MA Xing;MU Chunyang;JIANG Puzhao(College of Electronic&Information Engineering,North Minzu University,Yinchuan 750021,China;Key Laboratory of Intelligent Information and Big Data Processing of Ningxia Province North Minzu University,Yinchuan 750021,China;College of Mechatronic Engineering,North Minzu University,Yinchuan 750021,China)
出处
《传感器与微系统》
北大核心
2025年第5期133-137,共5页
Transducer and Microsystem Technologies
基金
银川市科技创新项目(2022GX04)
自治区科技创新领军人才培养工程项目(2021GKLRLX08)
宁夏回族自治区重点研发计划项目(2021BEE03002)。
关键词
卷积神经网络
深度可分离卷积
残差网络
抓取检测
convolutional neural network
depthwise separable convolution
residual network
grasping detection
