Research On Smoke Detection Algorithm Based On Deep Learning

The occurrence of fire will not only cause serious losses to economic property,but also bring certain hidden dangers to human life and safety.There will be a great deal of smoke in the early stage of a fire,but in the middle of the fire situation,there will be flames.Therefore,in order to prevent fires,it is very important to detect the smoke in time.Traditional smoke detection methods have disadvantages such as cumbersome feature design process,difficulty in extracting smoke features effectively,and excessive dependence on human priors for the quality of the model.In recent years,with the rapid development of computer technology and artificial intelligence,deep learning has attracted much attention.Many smoke detection algorithms related to deep learning have been proposed.Although they have good effect,many algorithms still have some problems,such as long time-consuming,high computational complexity and low detection accuracy.Aiming at the above problems,this paper proposes two smoke detection algorithms based on deep learning.Aiming at the problems of long time-consuming and high computational complexity of smoke detection,a smoke detection algorithm based on improved YOLOv3 is proposed in this paper.Firstly,depth-separable convolutions are introduced on the extraction layer of the Darknet-53 network.And depth-separable convolutions are used instead of standard convolutions to decrease the computation of the network.Introducing ECA into the three prediction layer branches of YOLOv3 network can quickly obtain the interaction information between cross channels.Then,CIo U Loss is used as the regression loss function of the prediction frame to effectively improve the regression stability of the prediction frame;Finally,Soft NMS is used to improve the detection accuracy of the network Experiments show that compared with Faster R-CNN,SSD,YOLOv3 and YOLOv4 algorithms,the improved YOLOv3 algorithm proposed in this paper can effectively improve the detection speed of smoke,and can be used in smoke scenes that require high real-time performance.Aiming at the problem of low detection accuracy of smoke detection algorithm,a smoke detection algorithm based on improved Retina Net is proposed in this paper.Firstly,the backbone network of Retina Net is replaced by Res Ne Xt to improve the detection accuracy of the network.Then,the attention module is introduced behind the convolutional layer of the Retina Net network to reduce the amount of network calculations and improve the detection accuracy of the network.Experiments show that compared with Faster R-CNN,SSD,YOLOv3,YOLOv4,Retina Net and improved YOLOv3 algorithms,the improved Retina Net algorithm proposed in this paper can effectively improve the detection accuracy of smoke,it can be applied to smoke scenes with high accuracy requirements.