Classification Of Hookah Images With Deep Learning

With the proliferation of social media usage, there has been a notable increase in posts related to hookah consumption, driven by its visually appealing presentation and distinctive consumption method. Such posts, especially in the context of the decreasing age of social media users, have the potential to exert an enticing influence. The potential impact of social media content on tobacco product consumption underscores the importance of identifying whether these posts contain tobacco-related imagery. Deep learning techniques have become increasingly prevalent in the detection of tobacco products within images and videos. However, existing studies have predominantly focused on cigarette detection due to its relatively higher consumption rates, leaving the area of hookah detection underexplored. In this study, the detection of hookah images in social media content was undertaken using the YOLO v5 algorithm, a version of convolutional neural networks, for the classification of hookah as an object in digital images. To achieve this, a dataset comprising 7200 hookah images of varying shapes, colors, and sizes was compiled from social media platforms and various e-commerce websites. The dataset was trained using the YOLO v5 xl model, and the experimental results indicated a detection accuracy of $98.88\%$. The findings of this experimental study indicate that the developed application could be employed to generate alerts or obscure hookah images in social media content. Consequently, this approach could contribute to raising awareness regarding the risks associated with hookah use and mitigating its potentially alluring effects.