Using Deep Neural Networks to Address the Evolving Challenges of Concealed Threat Detection within Complex Electronic Items

Abstract

X-ray baggage security screening is widely used to maintain aviation and transport safety and security. To address the future challenges of increasing volumes and complexities, the recent focus on the use of automated screening approaches are of particular interest. This includes the potential for automatic threat detection as a methodology for concealment detection within complex electronics and electrical items screened using low-cost, 2D X-ray imagery (single or multiple view). In this work, we use automatic object segmentation algorithms enabled by deep Convolutional Neural Networks (CNN, e.g. Mask R-CNN) together with the concept of image over-segmentation to the sub-component level and subsequently use CNN classification to determine the presence of anomalies at both an object or sub-component level. We evaluate the performance impact of three strategies: full frame, object segmentation, and object over-segmentation, for threat/anomaly detection within consumer electronics items. The experimental results exhibit that the object over-segmentation produces superior performance for the anomaly detection via classification, with $< 5%$ false positive and ~99% true positive.