Laboratory Safety Detection Of Hazardous Chemicals Based On Multi-sensor Fusion

The laboratory is an important guarantee for the cultivation of high-level talents,and has become the base of scientific and technological innovation and the foundation of social services.With the development of my country’s economy and the needs of all walks of life,in recent years,with the investment of the state and enterprises,the number of laboratories has increased significantly,providing a unique educational environment for my country’s scientific and technological innovation and the cultivation of highly streamlined talents,scientific research The strength of the technical team is also gradually increasing.But what follows is that these laboratory environments are becoming more and more complex and difficult to control;because of the expansion of business scale,the laboratory has more and more expensive instruments and equipment,chemical experiments A large number of hazardous chemicals are stored in the laboratory,and the safety problem of the laboratory has become increasingly prominent.In recent years,due to the occurrence of various laboratory accidents,laboratory safety issues have attracted widespread attention.In order to realize the safety supervision of the laboratory,the laboratory has adopted a series of safety management measures,but still due to the lack of perfect laboratory The safety supervision system still cannot realize the real-time and comprehensive safety guarantee of the laboratory.This topic aims at the safe operation requirements of hazardous chemical laboratories,based on target detection technology and supplemented by other sensors,a set of hazardous chemical laboratory safety detection system that can realize timely early warning function is established.This system has done a detailed study in order to realize the real-time early warning of the high incidence of fire accidents in the hazardous chemicals laboratory.The specific content is as follows: First,in order to effectively solve the problem that the smoke movement characteristics are not obvious during the smoldering stage of the fire,and the unbalanced positive and negative samples affected by the irrelevant background make the model difficult to converge,a method of modeling the mixed Gaussian background is proposed to eliminate the irrelevant background.Segment the suspected smoke movement area in the video image to improve the convergence speed and detection accuracy of the subsequent algorithm model.Second,in order to solve the problem that traditional sensors cannot provide real-time early warning of fire accidents under the limitation of large fire detection space distance,the safety early warning of the laboratory is improved by deploying and improving the YOLOv3 video target detection algorithm based on the video monitoring of hazardous chemical laboratories level.The specific improvement is divided into two parts: 1.In view of the problem that Dark Net-53 has too many convolutional layers and a large amount of parameters,which leads to excessive storage space and is difficult to deploy to monitoring equipment with limited resources,the lightweight network Efficient Net-B0 is improved.Replacing the original network Dark Net-53,the replaced network has fewer convolutional layers and reduces the amount of calculation through bottleneck flipping convolution,achieving a significant reduction in model storage space and network parameters,among which the application of the attention mechanism realizes the algorithm The model pays attention to the characteristics of smoke,and the characteristics of fire smoke have been extracted more effectively.2.In order to further reduce the impact of insufficient feature extraction after network replacement,a pooling fusion module is added between the backbone network and the FPN network,and the output shallow feature map is fused with the output deep feature map,enriching The feature expression information of the feature map greatly improves the detection accuracy of the algorithm model.Finally,tests were conducted on fire and non-fire smoke video datasets in various laboratory scenarios.The experimental results show that the improved detection algorithm in this paper can quickly and accurately identify fire smoke in hazardous chemicals laboratories,and the detection speed and accuracy Compared with the original YOLOv3 network,it has been improved by 2.1 times and 5.9%respectively.Subsequently,it was integrated and deployed into the produced fire smoke video monitoring equipment and combined with other sensors to realize the overall system.The system comprehensively integrated the information collected by various safety management in the laboratory,basically meeting the monitoring of laboratory environmental parameters,fire protection Real-time early warning and other requirements have effectively improved the laboratory’s safety assurance level and operational efficiency,and reduced safety management costs.