SMILE!

Social Media Image Labeling and Extraction (SMILE) -- Modern advancements in technology and connectivity have caused reconnaissance and intelligence gathering capabilities to evolve from traditional means, such as aerial and satellite photography, to cyber-based gathering. As social media becomes more prevalent in society, it creates a new platform to gather data and information, providing aid in intelligence acquisition. This project aims to answer the question, “How can convolutional neural networks and image recognition be used to further identify national security threats revealed through social media?” The main approach in this project uses Naive Bayesian statistics to predict whether an image pertains to national security based on the objects detected from a convolutional neural network. The final model scans through a set of images, identifies key features within those images, and labels images that contain data posing potential threats to national security with 85.5% accuracy. These results prove that conditioning the probability of an image being a threat on the objects detected in the convolutional neural networks improves overall model accuracy when searching for images of national security interest.