114Xray: A Large-Scale X-Ray Security Detection Benchmark and Aware Enhance Network for Real-World Prohibited Item Inspection in Baggage

Abstract

X-ray security inspection plays a crucial role in scenarios such as subway and express delivery. In the real world, prohibited items in baggage have abundant categories and complex morphology, resulting in unsatisfactory performance for detection. Moreover, the scarcity of high-quality datasets hinders the development of related research. In this work, we propose a large-scale multi-category X-ray security detection benchmark for real-world prohibited item inspection in baggage, named 114Xray. It consists of 1, 140, 000 X-ray images across 12 categories of prohibited items. Among them, 58, 000 images contain 75, 210 common prohibited items. Each image is obtained from real-world X-ray scans in express delivery and subway security inspection, and manually annotated through comprehensive careful examination by professional inspectors and algorithms, based on a self-developed annotation platform. To the best of our knowledge, it is the largest dataset in terms of the scale of data, morphology and category richness for prohibited item inspection. In addition, we propose an Aware Enhance Network (AENet) to handle the complex color distribution and diverse morphology of prohibited items on the 114Xray dataset, aiming to enhance the performance to perceive the material and edge of prohibited items. Extensive experiments validate the effectiveness of the 114Xray dataset and the superiority of AENet compared to the state-of-the-art methods. The 114Xray dataset is released at https://github.com/ming076/114Xray.