Threat Detection in Social Media Images Using the Inception-v3 Model

Abstract

Threat detection in images within social media content has become an important aspect of content monitoring. This task can be achieved using various image object detection and classification techniques. Recently, object detection has become an important task of machine learning, with significant studies dedicated to using deep learning techniques, especially the convolutional neural network (CNN), in the computer vision field. The current study involves an experiment using transfer learning technology to retrain the Inception-v3 model from TensorFlow in terms of the collected dataset related to known threat content. The novelty of this research work lies in the threat detection of images shared on social media which was not addressed before. The model achieves a high accuracy of around 96% in threat detection. The results of this research will be helpful in monitoring and tracking social media image content in terms of the detection of threats in the images shared among users, while the system can be used as a standalone system or as part of larger systems.