Real-Time YOLOv8l Contraband Detection in X- Ray Security Screening with Automated Identification of Weapons and Hazardous Objects for Threat Alerting

Abstract

A real-time contraband detection system based on YOLOv8l has been proposed for use in very high security installations, such as airports and public screening facilities. The contraband detection system has been trained with curated X-ray contraband datasets containing twelve different object types like firearms, knives, bullets, etc., with the datasets having been constructed to provide an equal number of examples for each object class to support robust detection. The methodology of building this automated object detection system includes the use of sophisticated data preprocessing, data augmentation, and the use of an anchorless detection framework provided by the YOLOv8l model, allowing for high accuracy via confidence thresholds and non-max suppression methods. In conjunction with the use of GPU acceleration for real-time inference, the system performs at high frame rates, making it appropriate for operational deployment in a security environment. The experimental evaluation results from the testing performed on the YOLOv8l based real-time contraband detection system demonstrated the performance success of the system as mAP@50 of 94.30%, indicating that the contraband detection system will produce very accurate detection results across all object classes. An audio alarm will be provided by the system to immediately alert security personnel to a detected high threat level item, thus improving the situational awareness for all security personnel engaged in contraband detection at security screening events. The results shown in this study validate that the proposed methodology will provide automated, accurate and efficient contraband detection in real-world security environments.