Research And Implementation Of Fire Detection Technology Based On Deep Learning

The emergence of fire has a decisive impact on the development and progress of human civilization.However,in daily life,the occurrence of fire will not only cause huge property losses,but also endanger life safety.Therefore,the in-depth study of fire detection technology has important theoretical significance and practical application value.The traditional fire detection technology based on sensors mainly uses sensors such as temperature and smoke to collect environmental information on site,and detects whether there is a fire by judging whether the detection value exceeds the alarm threshold.The application scope of this kind of fire detection technology is limited.The fire detection technology based on image processing uses different color space for feature matching and analysis,and integrates multiple features to judge whether the fire occurs or not.However,due to the complex visual features of the flame and irregular contour structure,the fire detection can only be realized for specific application scene,with high false detection and missing detection rate.Flame detection technology based on deep learning uses convolution neural network model to automatically extract feature information from the input image,and infers and predicts whether there is fire based on the feature image.However,the current network model for flame detection has complex structure,large parameter and model weight files,which requires high hardware storage resources.The main network does not make full use of the input image feature information,and the flame detection effect is poor in complex environment,and the small target flame is seriously missed.In addition,in practical application,the mode of collecting image at the edge end and returning it to the host end for detection is generally adopted,which has a large amount of data transmission and a certain delay.In some remote areas,the lack of 4G / 5G and other public wireless network platform also brings difficulties to the fire information return.In view of the above problems,this paper studies the fire detection technology,proposes a fire detection algorithm based on deep learning convolution neural network,and constructs a set of fire information remote monitoring system to realize the intelligent and automatic supervision of fire.This paper makes research and innovation in the following aspects.Based on convolution neural network,a fire detection network based on deep learning is proposed considering the parameters of network model and detection accuracy.Dense connection mechanism is used to design the main feature extraction network.In the dense layer,two channel convolution residual blocks are used to fuse different scale receptive field information,and attention module is introduced to adjust the channel weight of dense blocks.Based on the centernet network,the key points of the flame feature map are selected to predict the flame center point,and the flame width and height are regressed to obtain the fire detection results.The fire data set with various scenes and perfect labels is constructed.The accuracy of the fire detection network model proposed in this paper is fully verified through the evaluation index of the target detection model and the contrast experiment of the actual fire detection effect.The fire detection network model is deployed to the embedded system to realize the fire detection at the edge.According to the different application environment of indoor and outdoor,mountain forest and remote area,the dual channel fire information transmission scheme of Wi Fi wireless network and Beidou short message is designed and implemented,which transmits the edge end fire detection result image and fire detection information to the host.On the host side,the software platform of fire monitoring system is designed and implemented based on Django framework,and the fire information is managed and displayed in the browser page,so as to build a complete fire monitoring system.