A Deep Learning-based EfficientNetB0 Smoker Surveillance System for Enhancing Public Safety

Smoking is one of the risk factors associated with health conditions. The increase in smoking leads to various health issues such as lung cancer, oral cancer, respiratory problems, heart diseases and many more. Smoking doesn't only affect smokers but also the people in their surroundings. There are various places where, smoking is banned such as health institutes, hospitals, petrol pumps etc. However, due to limited resources of smoke identification in these places, people keep using various smoking materials such as cigarettes, and vaping. In a recent study, it has been analyzed that smoking can be one of the ignition sources in the petrol pump which can result in the complete burning of the pump and the entire damage to human health as well to the environment. The infammable vapors from petrol burning can cause many health issues to animals and human bodies resulting explosive atmosphere. In the proposed work, an EfficientNetB0 model is applied for the smoker and non-smoker classification, which can be applied to the places which require the monitoring of the smoke. This model has been fine-tuned with epoch values set to 5 and 10, learning rate values set to 0.001, and batch sizes set to 8 and 16. The model has been implemented with a balanced dataset of 1120 images, where the data resizing to 224X224 has been performed. Further, the outcomes have been analyzed in accuracy and loss, where the highest accuracy has been achieved as 92.78
^%
and the lowest loss of 0.1655 has been resulted. The proposed model can further be used in various other applications such as airlines, railway stations etc. which is the cost-effective and optimal solution to the problems caused by smoking habits.