Bidirectional Spatial Context Modeling and Contrast-Driven Feature Aggregation for Prohibited Item Detection in Cluttered X-Ray Baggage Images

Automated threat detection in X-ray security inspection is difficult. Objects often overlap in packed baggage. Many threats have very low contrast against surrounding items. As a result, detectors designed for natural images often miss small and occluded threats. This paper proposes the Scanline-Context X-ray Network (SCXNet), a detector designed for X-ray inspection. SCXNet uses a hierarchical encoder to preserve channel-specific features in shallow features and to build stronger representations in deeper features. It also uses Contrast-Driven Feature Aggregation to enhance weak boundaries through gradient-guided fusion. In addition, a Bidirectional Spatial Context Module aggregates context along the image-width axis with convolutional long short-term memory (ConvLSTM) to separate overlapping objects in clutter. SCXNet is trained with Varifocal Loss for classification and complete intersection over union (CIoU) loss for box regression. We evaluate SCXNet on two public datasets, Prohibited Item Detection Ray (PIDray) and Occluded Prohibited Items X-ray (OPIXray). SCXNet achieves higher mean average precision (mAP) than state-of-the-art detectors on both datasets. On PIDray, SCXNet reaches 73.48% mAP50 and improves over AO-DETR by 3.19% in relative terms. On OPIXray, it reaches 91.56% mAP50. These results show that domain tailored architecture design is effective for X-ray threat detection.

Authors

Publication Details

Journal
Algorithms
Published
2026-11-18
DOI
https://doi.org/10.3390/a19111016
Primary Topic
Advanced Neural Network Applications
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Bidirectional Spatial Context Modeling and Contrast-Driven Feature Aggregation for Prohibited Item Detection in Cluttered X-Ray Baggage Images

Xinmeng Wang, Hui Zhang
Algorithms
Advanced Neural Network Applications
article

Bidirectional Spatial Context Modeling and Contrast-Driven Feature Aggregation for Prohibited Item Detection in Cluttered X-Ray Baggage Images

Xinmeng Wang, Hui Zhang
article en

Abstract

Automated threat detection in X-ray security inspection is difficult. Objects often overlap in packed baggage. Many threats have very low contrast against surrounding items. As a result, detectors designed for natural images often miss small and occluded threats. This paper proposes the Scanline-Context X-ray Network (SCXNet), a detector designed for X-ray inspection. SCXNet uses a hierarchical encoder to preserve channel-specific features in shallow features and to build stronger representations in deeper features. It also uses Contrast-Driven Feature Aggregation to enhance weak boundaries through gradient-guided fusion. In addition, a Bidirectional Spatial Context Module aggregates context along the image-width axis with convolutional long short-term memory (ConvLSTM) to separate overlapping objects in clutter. SCXNet is trained with Varifocal Loss for classification and complete intersection over union (CIoU) loss for box regression. We evaluate SCXNet on two public datasets, Prohibited Item Detection Ray (PIDray) and Occluded Prohibited Items X-ray (OPIXray). SCXNet achieves higher mean average precision (mAP) than state-of-the-art detectors on both datasets. On PIDray, SCXNet reaches 73.48% mAP50 and improves over AO-DETR by 3.19% in relative terms. On OPIXray, it reaches 91.56% mAP50. These results show that domain tailored architecture design is effective for X-ray threat detection.

AlgorithmsVol. 19(11)
Fundamental Research Funds for the Central Universities
Advanced Neural Network Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon