ABSTRACT

Smoking poses significant health risks to both smokers and those exposed to secondhand smoke, and inappropriate smoking behavior can cause serious safety hazards such as fires or explosions, leading to substantial property damage. Therefore, enhancing the supervision of smoking bans in public places has become a critical concern, making the need for an intelligent smoking behavior detection system urgent. The core challenge lies in designing an algorithm capable of quickly and accurately detecting smoking behavior. With advancements in artificial intelligence, computer vision, deep learning, and hardware computing power, image processing technology has been increasingly applied in various fields, including face recognition, automation, and anomaly detection. Baidu has pioneered AI-based tobacco control by training a model using deep learning to automatically identify smokers in images. Building on this foundation, this paper proposes a smoking behavior detection algorithm based on deep learning image processing, which uses image data as input, with the deep learning network autonomously extracting target features and classifying them to detect smokers in images. This study analyzed and compared current deep learning-based target detection algorithms, focusing on accuracy and speed, developing a deep learning environment on a Win10 system using the YOLO algorithm, constructing a dataset, and training and evaluating the model. Based on the evaluation results, the YOLO structure is modified and parameters optimized to enhance performance.

Keywords: Artificial intelligence, computer vision, deep learning, smoking detection, YOLO algorithm
https://doi.org/10.57180/fyse2291