CigaretteCNN: A Convolutional Neural Network for Detecting Cigarette Smoking Activity

Abstract

Smoke detection has gained significant attention due to its implications for public health and safety. In this study, we proposed a method for detecting smoking activity using convolutional neural network (CNN). Our CNN architecture consists of four convolutional layers with different kernel sizes, followed by a max-pooling layer to achieve spatial down-sampling. The proposed model was trained and evaluated on a diverse dataset containing labeled instances of smoking and non-smoking activities. The training process uses stochastic gradient descent (SGD) with Nesterov momentum. The binary cross-entropy loss function was employed for optimization. We employed a validation set to mitigate early stopping and overfitting. Our experimental results demonstrate the effectiveness of the proposed CNN model for detecting smoking activity. The model achieved 89.51% accuracy on the test set which measures the model’s ability to accurately distinguish between smoking and non-smoking activities.