Multi-view Video Monitoring Method Of Fused Magnesia Melting Process

With the rapid development of video surveillance technology,video surveillance has brought much convenience in the monitoring of industrial production with its intuitive and visual performance.However,the video also has the defects of large amount of information and high storage capacity.How to utilize the video surveillance effectively in the industrial production process,and make fault diagnosis effectively from the industrial video information through the computer technology so as to identify the fault much timely,make the production process much safer has become an urgently issue to be researched.This thesis researches about industrial fused magnesia production and proposes a novel multi-view video summarization of fused magnesia production fault monitoring method and multi-view fused magnesia process fault monitoring method based on transfer learning,which provides a new idea for video monitoring of industrial production.The main contents of this thesis are as follows:(1)According to the idea of multi-view,this thesis proposes a monitoring method of fused magnesia process based on multi-view video summarization.The technology of multi-view video summarization extracts feature information from industrial video and compresses redundant video information to solve the problem that video information is complex and difficult to be processed,which has great advantages in fault detection and diagnosis.Multi-view video summarization of industrial production process can be obtained through the combination of shots segmentation,key frame extraction,similar shots clustering and shots optimization.It also takes into account the similarities and dissimilarities of content captured in different views,as well as the importance of the video content.This method is applied to the production process of fused magnesium furnace.The method can monitor and analyze the production process from multiple perspectives at the same time and improve the efficiency of fault monitoring in industrial production process.The experiment results also prove the effectiveness of the method.(2)In order to make efficient use of video data from different views,a multi-view video surveillance method based on transfer learning is proposed.Combined with the idea of transfer learning,this thesis finds an implicit public latent semantic subspace to bridge the source view and the target view,then it maps the data to this space to realize the cross-view processing monitor.At the same time,this thesis introduces the idea of graph regularization with taking into account the local characteristics of data,so that the different views can make good use of shared knowledge structure through transfer learning.The thesis is applied the method to the fault monitoring of the melting process of industrial fused magnesia furnace for the first time.It conforms the characteristics of modern industrial multi-view video surveillance.The experiment also proves that the method has some good fault monitoring results.