Smoking Behavior Detection Based On Deep Learning Image Processing

According to the statistics of the World Health Organization in 2019,there are 1.1 billion smokers in the world,of which the total number of smokers in China exceeds 360 million,making it the country with the largest number of smokers in the world.There are at least 70 kinds of carcinogens in tobacco.Smoking or passive smoking will cause great harm to human health.It costs about 3.5 trillion yuan to treat tobacco related diseases every year in China.Smoking is not only harmful to health,but also one of the fire hazards.About 20%of fires are caused by smoking every year.With the development of society,people pay close attention to the health,economic and environmental problems caused by smoking.Conventional smoking detection mainly uses smoke detector to determine whether there are people smoking in this area by detecting the cigarette smoke,but the accuracy of this method is low,it can only judge whether there are people smoking,and it can't accurately identify the smokers,and the smoke produced by smoking is less,so smoke detector is mostly used for fire alarm,and it can't judge whether there is smoking behavior.With the development of artificial intelligence,computer vision,deep learning and the improvement of hardware computing ability,image processing technology is gradually applied in many fields,such as face recognition,automatic driving,defect detection and so on.Baidu took the lead in proposing AI smoking control,using deep learning training model to automatically identify smokers in the image.Based on this background and foundation,this paper proposes a smoking behavior detection algorithm based on deep learning image processing,which takes the image data as input,extracts the target features and classifies the target independently through the deep learning network,so as to detect whether there is a smoker in the image.Through the analysis and comparison of the current target detection algorithm based on deep learning,combined with two major indicators of detection speed and accuracy,based on the YOLOv3 algorithm,a deep learning environment is built in Win10 system,data set is established,training and evaluation model are trained,and the structural improvement and parameter optimization of YOLOv3 are carried out according to the evaluation results.Theoretical analysis and experimental results show that the smoke detection algorithm based on YOLOv3 is feasible.The improved model has the detection speed of YOLOv3-tiny,and its detection accuracy is significantly higher than that of YOLOv3-tiny.At last,a smoking detection system is built,which uses PyQt5 to build the upper computer interface and calls the trained model to perform the smoking detection task.The test results show that the system can effectively identify the smokers in the image,and has a good promotion value.