Research And Application Of Fire Detection Based On Deep Learning

The traditional fire detection method uses a fire sensor.The fire alarm threshold is set in the MCU inside the fire sensor.As long as the threshold is exceeded,the MCU(Microcontroller Unit)will drive the internal alarm module to emit a continuous audible and visual alarm.A single fire sensor can only detect fire from a single angle,but the monitoring environment in which there are many interference factors for the fire sensor results in fire detection results that are often not accurate and reliable,and can no longer meet the needs of today’s fire detection.Therefore,the multi-sensor group is used to detect fire,which makes up for the shortcomings that a single fire sensor can only collect local fire characteristics,and uses the multi-sensor data fusion algorithm to perform fusion analysis on the collected fire data,which can effectively improve the fire detection ability.Fire detection by fire sensors is a contact detection.The detection range is limited,the detection time is long and the accuracy is not good.With the continuous development of computer vision,applying computer vision to fire detection can achieve more effective fire detection.The main work and results of this article are as follows:1.In view of the limitations of traditional single-type fire sensors in fire detection,multi-sensor groups are used for fire detection,and data fusion analysis is performed on the collected fire data through a multi-sensor data fusion algorithm.Aiming at the shortcomings of traditional data fusion algorithms,a DS-IMP(Dempster/Shafer-Improve)evidence theory model based on GRNN(Generalized Regression Neural Network)is proposed and compared with similar detection algorithms to verify the effectiveness of the model.Experimental results show that the model has good performance in fire detection and can achieve accurate and effective fire detection.2.Aiming at the shortcomings of fire detection,such as contact detection,limited detection range,long detection time,and poor accuracy of detection results,computer vision is applied to fire detection tasks,and a one-step target detection algorithm is used to detect fires.In view of the traditional YOLOV3(You Only Look Once Version3)’s poor ability to detect fires in small areas on fire detection tasks,YOLOV3-IMP is proposed for fire detection,and experiments are conducted on self-made data sets.Experimental results show that YOLOV3-IMP(You Only Look Once Version3-Improve)is more robust to fire detection than similar detection algorithms,and it has good results in both accuracy and speed.3.Designed and implemented a fire alarm system,which sends the fire images collected by video surveillance to the alarm platform,and uses the NB-IOT(Narrow Band Internet of Things)network to upload the fire data collected by NB smoke to the Internet of Things platform for unified management and Integration,and then upload the relevant data to the alarm platform for fire data analysis according to the needs of the alarm platform.The alarm platform processes the multi-sensor group information through the DS-IMP evidence theory model based on GRNN,and processes the image information through YOLOV3-IMP.Determine whether there is a fire,respond to the fire situation in real time,and realize remote monitoring,which can effectively protect the user’s property and life safety.