Research On Smoke And Flame Detection Based On Deep Learning

The occurrence of fire is a serious threat to the safety of human life and property.Therefore,the realization of fire detection is of great significance to the maintenance of people’s life and social stability.Traditional sensor and image processing methods have such problems as high cost,small scope of action,false alarm and delayed alarm,etc.In recent years,with the development of deep learning technology in the field of computer vision,the application of neural network model to smoke and flame detection has become a research hotspot.In view of the specific application of deep learning technology,this paper conducts research on pyrotechnic target detection in the following aspects:1.To solve the problem that there is no public dataset in pyrotechnic detection,this paper collected 18,417 pyrotechnic samples from baidu,google,git Hub,CSDN and other platforms,which provided a foundation for pyrotechnic detection to have a relatively high accuracy.2.In this paper,the current mainstream target detection algorithms Faster-R-CNN,SSD,YOLOV4 and YOLOV5 S are used to carry out pyrotechnic detection tasks.The experimental results show that YOLOV4 model has the highest MAP value of 77.8%,which is suitable for pyrotechnic detection tasks,except for FPS(FPS is only 8 frames per second,which does not meet the real-time requirements).3.In view of the large volume of YOLOV4 model,this paper proposes the Mobile Netv3-YOLOV4 model to quickly and accurately detect fireworks images.The Mobile Netv3 network is used to replace the feature extraction layer CSPDark Net53,and the deep separable convolution is used to replace the common convolution in PANET structure.In order to detect small target flame,a Densenet structural unit was embedded into the PANET structure of the model,and the attention mechanism ECANET was introduced to improve the accuracy.Finally,the number of parameters decreased by 43.4% when the MAP 2% was lost in the model.The model takes into account the detection speed and improves the detection ability of small target flame while maintaining a high detection accuracy,so it has more practical value.