摘要
为了提升电力生产中变电站安全监控的智能化水平,基于机器视觉理论对视频信号中的动作识别方法进行了研究,提出了一种基于空间通道、局部时域通道及全时域通道的三通道卷积神经网络(CNN)算法。该算法在上述的3个通道中分别使用视频信号的静态帧图像、光流图像与差分亮度信息作为CNN的输入,实现了视频序列在时域和空域上的关联,并有效提升了现有算法的学习能力。在对该网络参数进行识别时,为避免训练数据集冗余,还引入了基于Min-Batch思想的误差反向传播算法,从而有效提升了模型的泛化能力。仿真过程中,在每个通道中均使用了5个卷积层、3个池化层及2个全连接层的CNN网络。结果表明,在开放数据集上网络的识别精度较现有的Two-Stream CNN网络可提升3.09%;而在实际的生产数据集上,网络识别精度较Two-Stream CNN能够提升5.17%。
In order to improve the intelligent level of substation safety monitoring in power production,this paper studies the action recognition method in video signal based on machine vision theory,and extracts a three⁃channel Convolutional Neural Network(CNN)algorithm based on spatial channel,local time⁃domain channel and full time⁃domain channel.In the above three channels,the algorithm uses the static frame image,optical flow image and differential brightness information of video signal as the input of CNN,realizes the correlation of video sequence in time domain and space domain,and effectively improves the learning ability of existing algorithms.When identifying the network parameters,in order to avoid the redundancy of training data sets,this paper introduces an error back propagation algorithm based on Min⁃Batch idea,which effectively improves the generalization ability of the model.In the simulation process,the CNN network with 5 convolution layers,3 pooling layers and 2 full connection layers is used in each channel.The simulation results show that the recognition accuracy of the network can be improved by 3.09%compared with the existing Two⁃Stream CNN network on the open data set;On the actual production data set,the recognition accuracy of the network can be improved by 5.17% compared with Two⁃Streams CNN.
作者
翁凌雯
王栋
潘丹
谢乾武
胡东升
WENG Lingwen;WANG Dong;PAN Dan;XIE Qianwu;HU Dongsheng(State Grid Fujian Xintong Company,Fuzhou 350013,China;Anhui Nari Jiyuan Power Grid Technology Co.,Ltd.,Hefei 230601,China)
出处
《电子设计工程》
2023年第16期67-71,共5页
Electronic Design Engineering
基金
国家电网科技项目(2018BR3677)。
关键词
CNN
机器视觉
动作识别
变电站
安监管控
CNN
machine vision
action recognition
substation
safety supervision and control
