ORIGINAL RESEARCH article
Front. Phys.
Sec. Radiation Detectors and Imaging
Restoring Feature Reliability under X-ray Projection Degradation for Compact Prohibited-Item Detection
- BL
Bo Liu 1
- XL
Xiaoqun Liu 2
- YG
Yongqiang Gu 1
- GW
Guoping Wang 1
- AZ
Aihua Zhang 1
- ZX
Zhaoye Xing 2
1. Zhangjiakou Cigarette Factory Co., Ltd, Zhangjiakou, China
2. Hebei University of Architecture, Zhangjiakou, China
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Abstract
High-throughput X-ray security screening requires compact detectors that can keep reliable responses under projection overlap, material attenuation, weak boundaries, and dense baggage clutter. In practice, a baggage image is a two-dimensional radiation-attenuation projection of objects arranged at different depths; material superposition and boundary ambiguity can therefore corrupt target evidence before the detection head. This specific engineering problem is difficult for compact real-time detectors because the same model must preserve prohibited-item cues, suppress projection-induced interference, and satisfy latency and model-size constraints. We propose Aegis, a collaborative feature-reliability improvement method for compact X-ray prohibited-item detection. Aegis addresses projection-induced feature unreliability by jointly coordinating projection-aware local regulation, adaptive multi-level evidence aggregation, and response-preserving compact training within one deployable detector. On the Kaggle X-ray baggage benchmark, Aegis achieves 87.5% mAP@0.5 and 40.3% mAP@0.5:0.95 with 3.4 M parameters and 8.4 GFLOPs, improving YOLOv11s by 6.0 and 2.7 percentage points while using 64% fewer parameters. Under the projection degradation index (PDI), the gain increases from 2.2 percentage points in low-overlap images to 10.2 percentage points in heavy-overlap images. An operating-regime analysis further maps low, moderate, and severe projection degradation to sparse baggage projection, moderate-density projection, and dense superposed projection, indicating that the strongest benefit appears when overlap severity and projection density jointly increase. On PIDray, Aegis obtains 85.5%/64.5% mAP@0.5/mAP@0.5:0.95 at 5.1 ms latency. A collaborative reliability-improvement design thus improves compact X-ray security detection where projection degradation is most operationally critical.
Summary
Keywords
compact object detection, Feature reliability, Knowledge distillation, prohibited-item detection, projection degradation, radiation image analysis, X-ray security imaging
Received
30 May 2026
Accepted
14 August 2026
Copyright
© 2026 Liu, Liu, Gu, Wang, Zhang and Xing. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Xiaoqun Liu; Zhaoye Xing
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.