Research And Implementation Of Cigarette Detection Technology Based On Deep Learning

In order to avoid fire hazards caused by cigarettes and to safeguard environmental hygiene and cleanliness,tobacco control measures in nosmoking scenes should be further strictly implemented.Traditional tobacco control measures that rely on manual supervision and sensor detection are not only labour-intensive but also ineffective.With the development of target detection technology,the application of cigarette detection to actual no-smoking places,and the intelligent and automated detection of cigarettes,has practical implications for the strict implementation of tobacco control measures in relevant places.However,the existing deep learning-based cigarette detection technology has difficulties such as small size of cigarette targets,inconspicuous features and complex background,and the detection accuracy still needs to be improved.Therefore,this paper further investigates the problem of detecting small cigarette targets.In this paper,we propose a smoke detection algorithm based on Enhanced Feature Fusion(EFF)and attention mechanism.The algorithm is based on YOLOv5,and firstly,CBAM attention is introduced into the backbone network,so that the network can assign higher weights to the small target regions of the cigarette stubs during feature extraction.This enhances the network’s attention to the key features of small targets,avoiding the interference of complex background information and improving the feature extraction capability.An enhanced feature fusionbased neck network is then proposed,which fully fuses the local detailed features of the high-resolution feature map with the global semantic features of the low-resolution feature map by biaxially stitching three feature maps of different scales,solving the problem of insufficient characterization ability of small targets after multi-layer convolution.The final experimental results demonstrate that the proposed cigarette detection algorithm has a higher mAP50 than the original YOLOv5 algorithm,which improves the detection accuracy of small targets of cigarette stubs.Based on the above improved smoke detection algorithm,a smoke detection system is designed and implemented in this paper.The system has functional modules such as picture detection,video detection,real-time detection,review and evidence collection,and statistical analysis.The system detects smoke in uploaded images,uploaded videos and real-time monitoring,and alerts the relevant management if smoke is detected and forensically stored.The system also enables the querying of historical alarm data,reporting of false detections,calibration,statistical analysis and other operations.Finally,the system was tested and proved to be very accurate and real-time,which has practical significance and value for strictly implementing tobacco control measures in no-smoking scenes and ensuring the safety and hygiene of no-smoking places.