Research And Implementation Of Forest Fire Detection System Based On Deep Learning

In recent years,with the frequent occurrence of forest fires caused by the change of the earth's climate,it not only causes serious economic losses to the country,affects the ecological environment,but also poses a great threat to the safety of human life.Fire control,prevention first.If early warning is available,many fires can be put out before a large-scale disaster.In order to find forest fire quickly and accurately.At present,forest fire detection methods are mainly as follows: remote sensing image recognition by satellite generation,artificial watch tower,special patrol arrangement,patrol by aircraft or UAV.These methods have more or less different problems,such as the real-time and accuracy of satellite remote sensing monitoring is not high,the cost of watch tower and aircraft monitoring is too high,the workload of ground patrol is large and low efficiency,the sensors are easy to be interfered by the environment,and the accuracy is not high.Before the fire,a lot of smoke will be generated due to the combustion.The accurate detection of smoke detection can effectively prevent the fire.With the rapid development of computer vision technology,video image recognition and detection technology has been widely used in security,medical,intelligent agriculture,environmental protection and other aspects.This paper mainly uses computer vision technology to detect the smoke produced in the early stage of forest fire.Based on convolution neural network,two smoke detection algorithms are proposed,and the intelligent monitoring system of forest fire is designed by fusion detection algorithm.Finally,the advantages and disadvantages of the existing algorithms are summarized,and the corresponding solutions and further research directions are proposed.The main contents of this paper are as follows:1.At present,there is no unified data set for the research of forest fire smoke video.Through the collection and sorting in the preliminary preparation stage,21 pieces of forest fire smoke video for detecting the effect of algorithm and more than 40000 forest fire smoke pictures for training and testing convolution neural network are obtained.A set of smoke data set is established by normalized size and format.The video set contains many kinds of interference factors in the field forest environment,such as white clouds floating in the sky,fog,lake with similar smoke color,and shaking branches,which can better verify the effectiveness of the system,and provide availableresources for the research of forest fire smoke detection field.2.The color and dynamic characteristics of smoke are analyzed,and a smoke candidate area + Faster RCNN's smoke detection algorithm is proposed uses motion feature detection,color feature detection to extract smoke candidate area,and then combines convolution neural network to identify smoke image,which avoids the problem of single artificial selection of smoke feature with weak generalization,at the same time,the extraction of smoke candidate area reduces the calculation of neural network,making the algorithm more real-time and accurate Experiments show that the algorithm has higher accuracy than the traditional classifier.3.A Faster RCNN + C3 D smoke detection algorithm is proposed.Using Faster RCNN instead of traditional method to extract suspected smoke area,using Faster RCNN to extract static features,the RPN area recommendation network in Faster RCNN can get candidate areas with scores,and input candidate areas into C3 D network to obtain more dynamic features for classification.Compared with the algorithm of 2,Faster RCNN is used to extract the suspected smoke in the foreground,and C3 D is used to extract and classify the features.At the same time,compared with other algorithms of the same type,this algorithm has a higher detection accuracy,and also has a lower false alarm rate and missed detection rate.4.Based on the algorithm of 3,a set of forest fire intelligent monitoring system is designed,which mainly realizes the functions of video real-time browsing,forest environment detection,smoke recognition,fire alarm and so on.Through the experimental verification,the monitoring system can effectively achieve the goal of forest fire smoke monitoring.