A Novel Finetuned YOLOv8 Transfer Learning Model for Smoking Behavior Detection

This paper addresses the pressing need for smoking behavior detection in public health and safety monitoring, aiming to overcome the limitations of existing methods in real-time performance and accuracy, particularly in complex environments where precise monitoring is challenging. To this end, a transfer learning model based on YOLOv8 is proposed, where the pre-trained YOLOv8 architecture is fine-tuned to significantly enhance detection accuracy and real-time performance in specific scenarios, effectively adapting it to the task of smoking behavior detection. During training, data augmentation strategies such as random rotation, flipping, and brightness and contrast adjustments were employed to further optimize system performance. Experimental results demonstrate that the proposed model outperforms traditional methods in terms of accuracy, recall, and detection speed, and shows exceptional robustness under complex backgrounds and partial occlusions. This approach not only substantially improves indoor smoking detection but also holds great potential for widespread application in public spaces and complex environments.