Automated Identification of Hookahs (Waterpipes) on Instagram: An Application in Feature Extraction Using...

Area of Science:

• Public Health
• Computer Science
• Data Science

Background:

• Social media platforms like Instagram are valuable for public health surveillance but manual image analysis is inefficient.
• Existing automated image classification methods struggle with accurately distinguishing objects within images.

Purpose of the Study:

• To demonstrate the effectiveness of combining convolutional neural networks (CNN) for feature extraction with support vector machines (SVM) for image classification.
• To improve the accuracy of automated image analysis for public health surveillance using social media data.

Main Methods:

• Collected 840 images of waterpipes (hookah) from Instagram.
• Utilized a CNN to extract unique features from images containing waterpipes.
• Developed an SVM classifier to differentiate between images with and without waterpipes, comparing CNN+SVM to other methods.

Main Results:

• Increased validated training images and learned features by the SVM classifier led to higher accuracy.
• The CNN+SVM classifier achieved 99.5% accuracy in identifying hookah images.
• This combined approach outperformed methods using SVM, CNN, or bag-of-features alone.

Conclusions:

• Combining CNN for feature extraction with SVM for classification significantly enhances accuracy in image analysis.
• This method can expand the scope of image-based studies and monitor trends of emerging tobacco products on social media.
• The findings can inform public health policies and user experience research regarding emerging tobacco products.