Smoking Detection and Alarm in Multiple Scenes Based on Deep Learning

To achieve smoking detection and alarm in multiple scenes, reduce labor costs, and quickly identify smoking violations in public places from the urban electronic monitoring system, a method of smoking detection and alarm in multiple scenes based on deep learning YOLOv5 network model is proposed, and 4391 images of cigarettes and smokers are obtained by writing "Web crawler" program in Python. Divided into 3238 training sets and 1153 test sets, based on YOLOV51, YOLOV4 and YOLOV3 deep learning neural network models, group training experiments were designed to compare the training results. The results show that the recognition accuracy of the model based on the YOLOv51 model can reach 70. 8% on the test dataset, which has the accuracy of the current mainstream deep learning target detection algorithm, but it is superior to other target detection algorithms in terms of recognition speed and file volume. From the experimental results that can be drawn, YOLOv51 both recognition accuracy and recognition speed requirements, in dealing with multi-scene smoking detection and alarm has obvious advantages, and its greatly reduced model size is also suitable for public places monitoring system to rapid deploy, which can meet the actual needs of use well.