Melola Discovery

← Melola Discovery

Akses terbuka Jurnal SINTA 1

Real-Time Prohibited Item Detection in X-ray Security Screening via Adaptive Multi-scale Feature Fusion and Lightweight Dynamic Convolutions

Journal of Robotics and Control (JRC) · Vol. 6 No. 4 (2025) · hlm. 1746–1757

DOI
10.18196/jrc.v6i4.27030
ISSN
2715-5072, 2715-5056
Terbit
2025-06-30
Penerbit
Universitas Muhammadiyah Yogyakarta
Bahasa
eng
Lisensi
creativecommons.org/licenses/by-sa/4.0

Abstrak

Prohibited item detection in X-ray security screening is a challenging task due to the diverse shapes, sizes, and materials of concealed objects. In this paper, we propose a novel end-to-end framework, integrating adaptive multiscale convolution blocks (AMC Block) and an adaptive lightweight convolution module (ALCM), to address these challenges with high accuracy and efficiency. The AMC block leverages parallel convolutional paths with varying kernel sizes and dilation rates, enabling the capture of both fine-grained and large-scale features. This multiscale strategy ensures that small items like wires and larger objects such as bags or metallic weapons are equally well-detected. Building on top of multi-stage features extracted by the AMC block, we introduce the ALCM to refine and fuse feature maps at different pyramid levels. The ALCM employs a dynamic weight generator (DWG), which adaptively assigns importance to multiple convolutional kernels based on local content, followed by multi-scale depthwise convolutions (MSDC), a lightweight mechanism that enriches features across scales using parallel convolutions with different receptive fields. This approach enhances spatial context while keeping the parameter overhead minimal. Experimental results on two public large-scale X-ray datasets, OPIXray and HiXray, demonstrate that our method achieves state-of-the-art performance while maintaining real-time inference speed. Specifically, our model achieves 91.2% mAP@0.5 and 78.4% mAP@0.5:0.95 on OPIXray, and 87.3% mAP@0.5 and 73.5% mAP@0.5:0.95 on HiXray, outperforming strong baselines including YOLOv9 and Faster R-CNN. Despite competitive accuracy, our model remains efficient with 92.0 GFLOPs and 42 FPS. Furthermore, we examine the generalizability of our system across varied X-ray imaging settings and discuss failure cases such as false negatives in cluttered environments. These findings highlight the practical applicability of our approach for deployment in real-world security checkpoints, striking a strong balance between detection accuracy and computational efficiency.

Kata kunci

Jaringan kolaborasi para penulis

8 penulis · 17 tautan kepenulisan bersama

Arahkan kursor atau ketuk sebuah nama untuk menyorot rekan penulisnya. Klik dua kali untuk membuka halaman penulis.

  • Industrial University of Ho Chi Minh City
  • Afiliasi tidak diketahui
  • Organisasi berbeda
  • Organisasi sama
  • Afiliasi tidak diketahui
  • Garis lebih tebal = lebih banyak artikel bersama · lingkaran lebih besar = lebih banyak artikel
Tampilkan sebagai tabel
Rekan penulisOrganisasiArtikel bersama
Tuan Anh NguyenAfiliasi tidak diketahui3
Thanh Quyen NgoIndustrial University of Ho Chi Minh City2
Long Ho LeAfiliasi tidak diketahui2
Trung Dung NguyenIndustrial University of Ho Chi Minh City2
Trung Kien PhamAfiliasi tidak diketahui2
Phan Minh ThanIndustrial University of Ho Chi Minh City1

Afiliasi berasal dari OpenAlex; metadata OAI-PMH jurnal tidak memuatnya. Penulis tanpa ID OpenAlex dicocokkan berdasarkan nama, sehingga dua orang bisa tergabung atau satu orang bisa terpisah.

Sumber metadata: OAI-PMH (metadata jurnal) https://journal.umy.ac.id/index.php/jrc/oai, pertama di-harvest 9 Nov 2026, terakhir terlihat 16 Nov 2026.
Metadata asli (oai_dc)
<oai_dc:dc xmlns="http://www.openarchives.org/OAI/2.0/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:oai_dc="http://www.openarchives.org/OAI/2.0/oai_dc/" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://www.openarchives.org/OAI/2.0/oai_dc/  http://www.openarchives.org/OAI/2.0/oai_dc.xsd">
	<dc:title xml:lang="en-US">Real-Time Prohibited Item Detection in X-ray Security Screening via Adaptive Multi-scale Feature Fusion and Lightweight Dynamic Convolutions</dc:title>
	<dc:creator>Nguyen, Hoanh</dc:creator>
	<dc:creator>Ha, Chi Kien</dc:creator>
	<dc:subject xml:lang="en-US">Prohibited Item Detection</dc:subject>
	<dc:subject xml:lang="en-US">Real-Time X-Ray Screening</dc:subject>
	<dc:subject xml:lang="en-US">Adaptive Feature Fusion</dc:subject>
	<dc:subject xml:lang="en-US">Lightweight Dynamic Convolutions</dc:subject>
	<dc:subject xml:lang="en-US">Deployment-Efficient Vision Systems</dc:subject>
	<dc:description xml:lang="en-US">Prohibited item detection in X-ray security screening is a challenging task due to the diverse shapes, sizes, and materials of concealed objects. In this paper, we propose a novel end-to-end framework, integrating adaptive multiscale convolution blocks (AMC Block) and an adaptive lightweight convolution module (ALCM), to address these challenges with high accuracy and efficiency. The AMC block leverages parallel convolutional paths with varying kernel sizes and dilation rates, enabling the capture of both fine-grained and large-scale features. This multiscale strategy ensures that small items like wires and larger objects such as bags or metallic weapons are equally well-detected. Building on top of multi-stage features extracted by the AMC block, we introduce the ALCM to refine and fuse feature maps at different pyramid levels. The ALCM employs a dynamic weight generator (DWG), which adaptively assigns importance to multiple convolutional kernels based on local content, followed by multi-scale depthwise convolutions (MSDC), a lightweight mechanism that enriches features across scales using parallel convolutions with different receptive fields. This approach enhances spatial context while keeping the parameter overhead minimal. Experimental results on two public large-scale X-ray datasets, OPIXray and HiXray, demonstrate that our method achieves state-of-the-art performance while maintaining real-time inference speed. Specifically, our model achieves 91.2% mAP@0.5 and 78.4% mAP@0.5:0.95 on OPIXray, and 87.3% mAP@0.5 and 73.5% mAP@0.5:0.95 on HiXray, outperforming strong baselines including YOLOv9 and Faster R-CNN. Despite competitive accuracy, our model remains efficient with 92.0 GFLOPs and 42 FPS. Furthermore, we examine the generalizability of our system across varied X-ray imaging settings and discuss failure cases such as false negatives in cluttered environments. These findings highlight the practical applicability of our approach for deployment in real-world security checkpoints, striking a strong balance between detection accuracy and computational efficiency.</dc:description>
	<dc:publisher xml:lang="en-US">Universitas Muhammadiyah Yogyakarta</dc:publisher>
	<dc:date>2025-06-30</dc:date>
	<dc:type>info:eu-repo/semantics/article</dc:type>
	<dc:type>info:eu-repo/semantics/publishedVersion</dc:type>
	<dc:format>application/pdf</dc:format>
	<dc:identifier>https://journal.umy.ac.id/index.php/jrc/article/view/27030</dc:identifier>
	<dc:identifier>10.18196/jrc.v6i4.27030</dc:identifier>
	<dc:source xml:lang="en-US">Journal of Robotics and Control (JRC); Vol. 6 No. 4 (2025); 1746-1757</dc:source>
	<dc:source>2715-5072</dc:source>
	<dc:source>2715-5056</dc:source>
	<dc:language>eng</dc:language>
	<dc:relation>https://journal.umy.ac.id/index.php/jrc/article/view/27030/11690</dc:relation>
	<dc:rights xml:lang="en-US">Copyright (c) 2025 Hoanh Nguyen, Chi Kien Ha</dc:rights>
	<dc:rights xml:lang="en-US">https://creativecommons.org/licenses/by-sa/4.0</dc:rights>
</oai_dc:dc>