作者： He, Fangqiuzi;Liu, Yong;Zhan, Weiwen;Xu, Qingjie;Chen, Xiaoling

期刊： Energies,2022年15(6):2071 ISSN：1996-1073

通讯作者： Chen, XL

作者机构： [He, Fangqiuzi] Wuhan Polytech Univ, Sch Art & Design, Wuhan 430023, Peoples R China.;[He, Fangqiuzi; Liu, Yong; Zhan, Weiwen; Xu, Qingjie] China Univ Geosci, Sch Mech Engn & Elect Informat, Wuhan 430074, Peoples R China.;[Chen, Xiaoling] China Univ Geosci, Sch Art & Media, Wuhan 430074, Peoples R China.

通讯机构： [Chen, XL ] C;China Univ Geosci, Sch Art & Media, Wuhan 430074, Peoples R China.

关键词： Graph convolutional neural network;Manual operation accuracy evaluation;Virtual reality

摘要： The standard of manual operation in smart grid, which require accurate manipulation, is high, especially in experimental, practice, and training systems based on virtual reality (VR). In the VR training system, data gloves are often used to obtain the accurate dataset of hand movements. Previous works rarely considered the multi-sensor datasets, which collected from the data gloves, to complete the action evaluation of VR training systems. In this paper, a vectorized graph convolutional deep learning model is proposed to evaluate the accuracy of test actions. First, the kernel of vectorized spatio-temporal graph convolutional of the data glove is constructed with different weights for different finger joints, and the data dimensionality reduction is also achieved. Then, different evaluation strategies are proposed for different actions. Finally, a convolution deep learning network for vectorized spatio-temporal graph is built to obtain the similarity between test actions and standard ones. The evaluation results of the proposed algorithm are compared with the subjective ones labeled by experts. The experimental results verify that the proposed action evaluation method based on the vectorized spatio-temporal graph convolutional is efficient for the manual operation accuracy evaluation in VR training systems of smart grids. © 2022 by the authors. Licensee MDPI, Basel, Switzerland.

语种： 英文