Research On The Application Of Smart Fire Protection Based On Deep Learning

Artificial intelligence technologies such as path planning,target detection and behavior recognition are important research contents of today’s smart fire protection.Traditional path planning,target detection and behavior recognition technologies have also been successfully applied in various fields,however,applying the above traditional technologies to firefighting scenarios mainly has the following shortcomings:due to the increasing complexity of the building structure,when traditional path planning is used for fire rescue guidance,it cannot adapt to complex and changeable fire scenarios and cannot grasp detailed fire information in real time;Traditional target detection and behavior recognition technologies only focus on people or objects themselves and cannot obtain related information between people and objects,and thus cannot early warning of fires caused by man-made causes.In response to the above two major issues,This paper proposes an agent rescure path planning algorithm based on deep reinforcement learning and a fire hazard recognition algorithm based on deep learning,and apply these two algorithms to the enterprise level in the smart fire management system,the functions of fire warning and fire rescue guidance are realized.Specifically,the main innovations and research work of this paper are as follows:1.Aiming at the problems existing in the application of current path planning algorithms to fire rescue,this paper proposes an agent path planning algorithm based on deep reinforcement learning.The algorithm uses Deep Q Network(DQN)as the basic network to provide solutions for the perception and decision-making of agents in complex environments;at the same time,it uses convolutional neural networks to identify fire points.When a fire point is detected,The intelligent body transmits the location of the fire point and the learned optimal rescue route to outside rescuers to realize the intelligent fire rescue guidance function.Experiments show that the algorithm proposed in this paper can not only provide the optimal route from the safe escape port to the location of each fire point,but also efficiently detect the fire point and location information in the scene.2.Aiming at the problem that the current behavior recognition algorithm only focuses on the human body and cannot correlate the interaction behavior between people and things,This paper proposes a human-object interaction behavior recognition algorithm based on deep learning to solve the fire hazards caused by human causes.The target detection algorithm based on YOLO V3 and the bone extraction algorithm based on OpenPose are merged.By constructing a behavior knowledge base for matching,the behavior recognition when people are associated with objects can be recognized,and behaviors such as whether people smoke,whether they hold lighters,cigarettes,etc.Identify it.Experiments show that the algorithm proposed in this paper has an accuracy rate of 91.3%for the recognition of smoking behavior,has good usability and reliability,and meets the requirements of smart fire warning applications.3.Based on the two algorithms proposed in this paper,research the application of algorithms in smart fire management systems.Focus on the supervision of smoking behavior in public places,rescue guidance in the event of a fire,fire equipment inspection and maintenance,fire data visualization,fire information query and other functions,to achieve the intelligent fire management goal of "prevention first,combination of prevention and fire prevention" to protect the enterprise production safety.Figure[47]table[10]reference[72]...