pith. sign in

arxiv: 2605.20669 · v1 · pith:AFWYDOAOnew · submitted 2026-05-20 · 💻 cs.CV

GSA-YOLO: A High-Efficiency Framework via Structured Sparsity and Adaptive Knowledge Distillation for Real-Time X-ray Security Inspection

Jiahao Kong This is my paper

Pith reviewed 2026-05-21 05:42 UTC · model grok-4.3

classification 💻 cs.CV
keywords X-ray security inspectionreal-time object detectionstructured sparsityknowledge distillationYOLOlightweight modelprohibited item detection
0
0 comments X

The pith

GSA-YOLO combines Group Lasso sparsity, head pruning, and adaptive distillation on YOLOv8n to raise X-ray detection accuracy at higher speed.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper sets out to solve the tension between speed and accuracy in real-time X-ray security scanning where items are often occluded or surrounded by clutter. It starts from the YOLOv8n detector and adds three coordinated changes: Group Lasso regularization on the neck to strengthen feature extraction, Sparse Structure Selection on the detection head to remove redundant channels, and an Adaptive Knowledge Distillation stage that transfers knowledge from a larger teacher to restore any lost accuracy. The authors show that this combination cuts FLOPs from 8.7 G to 8.0 G while lifting mAP50:95 by 2.4 percent on HiXray and 1.8 percent on PIDray and reaching 189.62 frames per second. A sympathetic reader would care because the result points toward detectors that can run on modest hardware yet still catch prohibited objects reliably in live inspection lines.

Core claim

GSA-YOLO integrates Group Lasso on the neck for robust feature extraction, Sparse Structure Selection on the head for model slimming, and Adaptive Knowledge Distillation for accuracy recovery, yielding 189.62 FPS, a drop in computational cost from 8.7 G to 8.0 G, and mAP50:95 scores of 0.531 on HiXray and 0.679 on PIDray, each an improvement over the unmodified YOLOv8n baseline.

What carries the argument

Structured sparsity produced by Group Lasso applied to the neck together with Sparse Structure Selection on the head, followed by Adaptive Knowledge Distillation to restore performance on the YOLOv8n backbone.

If this is right

  • X-ray security systems could process streams at nearly 190 frames per second on standard hardware while using fewer floating-point operations.
  • The same pruning-plus-distillation recipe could be applied to other single-stage detectors when both latency and accuracy matter.
  • Detection heads can be slimmed substantially without permanent accuracy loss if the distillation step is adapted to the remaining channels.
  • Real-time prohibited-item alerts become feasible on edge devices placed at checkpoints.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method might extend to other imaging modalities that share heavy occlusion and strict speed limits, such as baggage CT or industrial defect scanning.
  • Further channel pruning could be explored by varying the Group Lasso penalty strength across different neck stages rather than applying it uniformly.
  • A direct comparison against quantization-aware training would clarify whether sparsity or quantization delivers the larger efficiency gain for this task.

Load-bearing premise

The chosen mix of Group Lasso, Sparse Structure Selection, and Adaptive Knowledge Distillation will keep or improve robustness when the X-ray images contain occlusion levels, clutter patterns, or scanner hardware different from those in the HiXray and PIDray collections.

What would settle it

Running the model on a new X-ray test set collected from a different scanner model that contains higher average occlusion and reporting either lower mAP50:95 than the baseline or slower inference would falsify the central performance claim.

Figures

Figures reproduced from arXiv: 2605.20669 by Jiahao Kong.

Figure 1
Figure 1. Figure 1: GSA-YOLO framework overview based on YOLOv8n for X-ray security inspec [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The detailed illustration of SSS module. SSS performs channel pruning based [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The detailed illustration of Ada-KD mechanism. The total loss function used during the Ada-KD phase is defined as: Ltotal_Ada = Ldet + λ0 · α (t) LKD (8) where Ltotal_Ada represents the overall loss during the distillation phase. Ldet is the base detection loss calculated using the ground-truth hard labels; LKD is the knowledge distillation loss, which is computed using the soft labels provided by the teac… view at source ↗
Figure 4
Figure 4. Figure 4: The trade-off analysis of teacher model complexity (GFLOPs) vs. student model [PITH_FULL_IMAGE:figures/full_fig_p022_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Hyperparameter sensitivity analysis for the GSA-YOLO framework. The shaded [PITH_FULL_IMAGE:figures/full_fig_p024_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Confusion matrix plots on the HiXray and PIDray hidden test sets. Subfigures [PITH_FULL_IMAGE:figures/full_fig_p029_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Class Activation Mapping visualization comparison for X-ray prohibited item [PITH_FULL_IMAGE:figures/full_fig_p031_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: X-ray security inspection detection visualization comparison. The figure shows [PITH_FULL_IMAGE:figures/full_fig_p032_8.png] view at source ↗
read the original abstract

X-ray security inspection requires accurate real-time detection of prohibited items, but existing models often struggle to balance the challenges of severe occlusion, complex clutter, and strict speed requirements. To overcome these challenges, this paper proposes GSA-YOLO, a novel lightweight framework built upon the YOLOv8n architecture, specifically engineered to enhance detection robustness and inference efficiency. GSA-YOLO strategically integrates structured sparsity and adaptive knowledge transfer through three core components: Group Lasso (GL) applied to the network neck for robust feature extraction; Sparse Structure Selection (SSS) applied to the detection head for significant model slimming; and an Adaptive Knowledge Distillation (Ada-KD) mechanism for comprehensive accuracy recovery. This integrated approach synergistically enhances feature representation while pruning redundant channels, maximizing model efficiency without sacrificing performance. Rigorous evaluations on the HiXray and PIDray datasets confirm GSA-YOLO's comprehensive capability, achieving a leading inference speed of 189.62 FPS, accompanied by a reduction in computational cost from 8.7G to 8.0G. Crucially, GSA-YOLO secures mAP50:95 results of 0.531 and 0.679 on HiXray and PIDray, demonstrating 2.4% and 1.8% improvements over the baseline, respectively. Compared to other models, GSA-YOLO exhibits enhanced accuracy while maintaining computational efficiency, making it a promising solution for practical X-ray security inspection.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes GSA-YOLO, a lightweight framework extending YOLOv8n for real-time X-ray security inspection. It applies Group Lasso regularization to the neck for robust feature extraction, Sparse Structure Selection to the detection head for model slimming, and Adaptive Knowledge Distillation to recover accuracy after pruning. On the HiXray and PIDray datasets, GSA-YOLO reports 189.62 FPS, FLOPs reduced from 8.7G to 8.0G, and mAP50:95 scores of 0.531 and 0.679 (improvements of 2.4% and 1.8% over the YOLOv8n baseline).

Significance. If the performance deltas are shown to arise specifically from the GL+SSS+Ada-KD pipeline under matched training conditions, the work supplies a concrete efficiency improvement for occlusion-heavy X-ray detection tasks. The reported speed-accuracy operating point is relevant for deployment in security screening pipelines.

major comments (2)
  1. [§4 Experiments and Table 2] §4 Experiments and Table 2: The 2.4% and 1.8% mAP50:95 gains over YOLOv8n are reported without stating that the baseline was retrained under identical optimizer, scheduler, augmentation, epoch count, and loss weighting. Because the FLOPs reduction is modest (8.7G to 8.0G), small hyper-parameter differences alone could produce the observed trade-off; explicit confirmation of matched training protocols is required to attribute the gains to the proposed components.
  2. [§3.2 and §3.3] §3.2 and §3.3: The Adaptive Knowledge Distillation mechanism is described at a high level but lacks the explicit loss formulation or schedule for the adaptation weights/temperature. Without these details, it is impossible to verify how accuracy is recovered after the Group Lasso and Sparse Structure Selection steps or to reproduce the exact contribution of Ada-KD.
minor comments (2)
  1. [§4] The results tables should report standard deviations or error bars across multiple random seeds for both mAP and FPS to allow assessment of statistical significance of the reported improvements.
  2. [§3.1] Implementation details for the Group Lasso regularization strength and Sparse Structure Selection thresholds should be listed explicitly (values, selection criteria) to support reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments that highlight important aspects of reproducibility and clarity. We address each major comment below and will revise the manuscript accordingly to strengthen the presentation of our results.

read point-by-point responses

  1. Referee: [§4 Experiments and Table 2] §4 Experiments and Table 2: The 2.4% and 1.8% mAP50:95 gains over YOLOv8n are reported without stating that the baseline was retrained under identical optimizer, scheduler, augmentation, epoch count, and loss weighting. Because the FLOPs reduction is modest (8.7G to 8.0G), small hyper-parameter differences alone could produce the observed trade-off; explicit confirmation of matched training protocols is required to attribute the gains to the proposed components.

    Authors: We agree that explicit confirmation of matched training conditions is essential to attribute the observed gains specifically to the GL+SSS+Ada-KD pipeline. The YOLOv8n baseline was retrained from scratch under identical conditions to GSA-YOLO, including the same optimizer, learning-rate scheduler, augmentation pipeline, epoch count, batch size, and loss weighting. We will add a dedicated paragraph in §4 clarifying these matched protocols and will update the caption of Table 2 to state that all reported results use the same training configuration. revision: yes

  2. Referee: [§3.2 and §3.3] §3.2 and §3.3: The Adaptive Knowledge Distillation mechanism is described at a high level but lacks the explicit loss formulation or schedule for the adaptation weights/temperature. Without these details, it is impossible to verify how accuracy is recovered after the Group Lasso and Sparse Structure Selection steps or to reproduce the exact contribution of Ada-KD.

    Authors: We acknowledge that the current description of Ada-KD remains at a conceptual level. To enable verification and reproduction, we will expand §3.2 and §3.3 with the precise loss formulation (a weighted combination of detection loss and distillation loss), the functional form of the adaptive temperature schedule, and the rule used to adjust the adaptation weights during training. These additions will be accompanied by the corresponding equations and a brief pseudocode snippet. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical measurements on fixed external datasets

full rationale

The paper proposes GSA-YOLO by combining Group Lasso on the neck, Sparse Structure Selection on the head, and Adaptive Knowledge Distillation for accuracy recovery on top of YOLOv8n. All reported outcomes (189.62 FPS, 8.7G to 8.0G FLOPs, mAP50:95 of 0.531/0.679 with +2.4%/+1.8% gains) are presented as direct measurements on the independent HiXray and PIDray datasets. No equations, parameters, or predictions are defined in terms of the target metrics themselves, no fitted inputs are relabeled as predictions, and no load-bearing steps reduce to self-citations or ansatzes imported from prior author work. The derivation chain consists of architectural modifications followed by standard training and evaluation; results remain falsifiable against external benchmarks and do not collapse to internal fits by construction.

Axiom & Free-Parameter Ledger

3 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard deep-learning assumptions about the base YOLOv8n architecture and the effectiveness of sparsity and distillation for this domain; no new entities are postulated.

free parameters (3)
  • Group Lasso regularization strength
    Tuned hyperparameter controlling sparsity in the neck; value not stated in abstract.
  • Sparse Structure Selection thresholds
    Parameters deciding which channels or structures to prune in the detection head.
  • Adaptive Knowledge Distillation weights or temperature
    Parameters governing how teacher knowledge is transferred to recover accuracy.
axioms (2)
  • domain assumption YOLOv8n is an appropriate starting architecture for X-ray prohibited-item detection
    The framework is built directly upon YOLOv8n without justifying why this base is optimal for the occlusion and clutter characteristics of X-ray images.
  • domain assumption HiXray and PIDray datasets sufficiently represent real security inspection conditions
    Performance claims are evaluated only on these two datasets; generalization is assumed.

pith-pipeline@v0.9.0 · 5794 in / 1549 out tokens · 35411 ms · 2026-05-21T05:42:57.085996+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

59 extracted references · 59 canonical work pages · 5 internal anchors

  1. [1]

    Kernel PCA for novelty detection

    S. Akcay, T. Breckon, Towards automatic threat detection: a survey of advances of deep learning within X-ray security imaging, Pattern Recog- nition 122 (2022) 108245.doi:https://doi.org/10.1016/j.patcog. 2021.108245

    work page doi:10.1016/j.patcog 2022

  2. [2]

    J. Wu, X. Xu, J. Yang, Object detection and x-ray security imaging: A survey, IEEE Access 11 (2023) 45416–45441.doi:https://doi.org/ 10.1109/access.2023.3273736

    work page doi:10.1109/access.2023.3273736 2023

  3. [3]

    unexpected item in the bagging area

    L. D. Griffin, M. Caldwell, J. T. Andrews, H. Bohler, “unexpected item in the bagging area”: Anomaly detection in x-ray security images, IEEE Transactions on Information Forensics and Security 14 (6) (2018) 1539– 1553.doi:https://doi.org/10.1109/tifs.2018.2881700

    work page doi:10.1109/tifs.2018.2881700 2018

  4. [4]

    Z. Wang, X. Wang, Y. Shi, H. Qi, M. Jia, W. Wang, Lightweight detec- tion method for x-ray security inspection with occlusion, Sensors 24 (3) (2024) 1002.doi:https://doi.org/10.3390/s24031002

    work page doi:10.3390/s24031002 2024

  5. [5]

    Y. Li, S. Li, H. Du, L. Chen, D. Zhang, Y. Li, Yolo-acn: Focus- ing on small target and occluded object detection, IEEE access 8 (2020) 227288–227303.doi:https://doi.org/10.1109/access.2020. 3046515. 34

    work page doi:10.1109/access.2020 2020

  6. [6]

    Laroca, E

    R. Laroca, E. Severo, L. A. Zanlorensi, L. S. Oliveira, G. R. Gonçalves, W. R. Schwartz, D. Menotti, A robust real-time automatic license plate recognition based on the yolo detector, in: 2018 international joint conference on neural networks (ijcnn), IEEE, 2018, pp. 1–10. doi:https://doi.org/10.1109/ijcnn.2018.8489629

    work page doi:10.1109/ijcnn.2018.8489629 2018

  7. [7]

    T. Shao, D. Shin, Structured pruning for deep convolutional neural networks via adaptive sparsity regularization, in: 2022 IEEE 46th Annual Computers, Software, and Applications Conference (COMP- SAC), IEEE, 2022, p. 982–987.doi:https://doi.org/10.1109/ compsac54236.2022.00151. URLhttp://dx.doi.org/10.1109/compsac54236.2022.00151

    work page doi:10.1109/compsac54236.2022.00151 2022

  8. [8]

    Y. Li, S. Gu, C. Mayer, L. Van Gool, R. Timofte, Group sparsity: the hinge between filter pruning and decomposition for network compres- sion, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020, pp. 8018–8027.doi:https: //doi.org/10.1109/cvpr42600.2020.00804

    work page doi:10.1109/cvpr42600.2020.00804 2020

  9. [9]

    Huang, N

    Z. Huang, N. Wang, Data-driven sparse structure selection for deep neural networks, in: V. Ferrari, M. Hebert, C. Sminchisescu, Y. Weiss (Eds.), Computer Vision – ECCV 2018, Vol. 11220, Springer Interna- tional Publishing, Cham, 2018, pp. 317–334.doi:https://doi.org/ 10.1007/978-3-030-01270-0_19

    work page doi:10.1007/978-3-030-01270-0_19 2018

  10. [10]

    Distilling the Knowledge in a Neural Network

    G. Hinton, O. Vinyals, J. Dean, Distilling the knowledge in a neural network (2015).doi:https://doi.org/10.48550/arXiv.1503.02531

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1503.02531 2015

  11. [11]

    Paying More Attention to Attention: Improving the Performance of Convolutional Neural Networks via Attention Transfer

    S. Zagoruyko, N. Komodakis, Paying more attention to attention: im- proving the performance of convolutional neural networks via attention transfer (2017).doi:https://doi.org/10.48550/arXiv.1612.03928

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1612.03928 2017

  12. [12]

    Boutros, V

    F. Boutros, V. Štruc, N. Damer, Adadistill: adaptive knowledge distil- lation for deep face recognition, in: Computer Vision – ECCV 2024, Springer Nature Switzerland, 2025, pp. 163–182.doi:https://doi. org/10.1007/978-3-031-73001-6_10

    work page doi:10.1007/978-3-031-73001-6_10 2024

  13. [13]

    Z. Zhu, Y. Zhu, H. Wang, N. Wang, J. Ye, X. Ling, Fdtnet: Enhancing frequency-aware representation for prohibited object detection from x- ray images via dual-stream transformers, Engineering Applications of 35 Artificial Intelligence 133 (2024) 108076.doi:https://doi.org/10. 1016/j.engappai.2024.108076

    work page arXiv 2024

  14. [14]

    J. Ding, C. Ye, H. Wang, J. Huyan, M. Yang, W. Li, Foreign bodies detector based on detr for high-resolution x-ray images of textiles, IEEE TransactionsonInstrumentationandMeasurement72(2023)1–10.doi: https://doi.org/10.1109/TIM.2023.3246510

    work page doi:10.1109/tim.2023.3246510 2023

  15. [15]

    Y. Wei, H. Wang, Cooperative distillation with x-ray images classi- fiers for prohibited items detection, Engineering Applications of Arti- ficial Intelligence 127 (2024) 107276.doi:https://doi.org/10.1016/ j.engappai.2023.107276

    work page arXiv 2024

  16. [16]

    Chang, Y

    A. Chang, Y. Zhang, S. Zhang, L. Zhong, L. Zhang, Detecting pro- hibited objects with physical size constraint from cluttered x-ray bag- gage images, Knowledge-Based Systems 237 (2022) 107916.doi:https: //doi.org/10.1016/j.knosys.2021.107916

    work page doi:10.1016/j.knosys.2021.107916 2022

  17. [17]

    Hassan, S

    T. Hassan, S. Akcay, M. Bennamoun, S. Khan, N. Werghi, A novel in- cremental learning driven instance segmentation framework to recognize highly cluttered instances of the contraband items, IEEE Transactions on Systems, Man, and Cybernetics: Systems 52 (11) (2021) 6937–6951. doi:https://doi.org/10.1109/TSMC.2021.3131421

    work page doi:10.1109/tsmc.2021.3131421 2021

  18. [18]

    C. Ma, L. Zhuo, J. Li, Y. Zhang, J. Zhang, Eaod-net: Effective anomaly object detection networks for x-ray images, IET Image Process- ing 16 (10) (2022) 2638–2651.doi:https://doi.org/10.1049/ipr2. 12514

    work page doi:10.1049/ipr2 2022

  19. [19]

    Y. Wei, Y. Wang, H. Song, Cfpa-net: cross-layer feature fusion and parallel attention network for detection and classification of prohibited items in x-ray baggage images, in: 2021 IEEE 7th International Con- ference on Cloud Computing and Intelligent Systems (CCIS), IEEE, 2021, pp. 203–207.doi:https://doi.org/10.1109/CCIS53392.2021. 9754631

    work page doi:10.1109/ccis53392.2021 2021

  20. [20]

    G. D. Raj, B. Prabadevi, Enhancing surface detection: a comprehensive analysis of various YOLO models, Heliyon 11 (3) (2025).doi:https: //doi.org/10.1016/j.heliyon.2025.e42433. 36

    work page doi:10.1016/j.heliyon.2025.e42433 2025

  21. [21]

    Y. Zhou, X. Xu, R. Wang, Ei-YOLO: efficiently improved YOLO on detection of prohibited items during security inspections, in: Pattern Recognition and Computer Vision, Springer Nature, Singapore, 2025, pp. 330–343.doi:https://doi.org/10.1007/978-981-97-8858-3_ 23

    work page doi:10.1007/978-981-97-8858-3_ 2025

  22. [22]

    A. Wang, P. Yuan, H. Wu, Y. Iwahori, Y. Liu, Improved YOLOv8 for dangerous goods detection in X-ray security images, Electronics 13 (16) (2024) 3238.doi:https://doi.org/10.3390/electronics13163238

    work page doi:10.3390/electronics13163238 2024

  23. [23]

    LeCun, J

    Y. LeCun, J. Denker, S. Solla, Optimal brain damage, in: Advances in Neural Information Processing Systems, Vol. 2, Morgan-Kaufmann, 1989, pp. 589–605.doi:https://doi.org/10.3156/jfuzzy.9.1_50_2

    work page doi:10.3156/jfuzzy.9.1_50_2 1989

  24. [24]

    S. Han, H. Mao, W. J. Dally, Deep compression: compressing deep neural networks with pruning, trained quantization and huffman coding (2016).doi:10.48550/arXiv.1510.00149

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1510.00149 2016

  25. [25]

    Z. Liu, J. Li, Z. Shen, G. Huang, S. Yan, C. Zhang, Learning efficient convolutional networks through network slimming, in: Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2017, pp. 2736–2744.doi:https://doi.org/10.1109/iccv.2017.298

    work page doi:10.1109/iccv.2017.298 2017

  26. [26]

    M. Yuan, Y. Lin, Model selection and estimation in regression with grouped variables, Journal of the Royal Statistical Society Series B (Sta- tistical Methodology) 68 (1) (2006) 49–67.doi:https://doi.org/10. 1111/j.1467-9868.2005.00532.x

    work page arXiv 2006

  27. [27]

    K. Bui, F. Park, S. Zhang, Y. Qi, J. Xin, Structured sparsity of con- volutional neural networks via nonconvex sparse group regularization, Frontiers in Applied Mathematics and Statistics 6 (2021).doi:https: //doi.org/10.3389/fams.2020.529564

    work page doi:10.3389/fams.2020.529564 2021

  28. [28]

    Y. Shao, K. Zhao, Z. Cao, Z. Peng, X. Peng, P. Li, Y. Wang, J. Ma, Mobileprune: neural network compression viaℓ0 sparse group lasso on the mobile system, Sensors 22 (11) (2022) 4081.doi:https://doi. org/10.3390/s22114081

    work page doi:10.3390/s22114081 2022

  29. [29]

    M. Park, D. Kim, C. Park, Y. Park, G. E. Gong, W. W. Ro, S. Kim, Reprune: channel pruning via kernel representative selection, Proceed- 37 ings of the AAAI Conference on Artificial Intelligence 38 (13) (2024) 14545–14553.doi:https://doi.org/10.1609/aaai.v38i13.29370

    work page doi:10.1609/aaai.v38i13.29370 2024

  30. [30]

    M. Lin, R. Ji, Y. Zhang, B. Zhang, Y. Wu, Y. Tian, Channel pruning via automatic structure search (2020).doi:https://doi.org/10.48550/ arXiv.2001.08565

    work page arXiv 2020

  31. [31]

    Bae, Y .-J

    H. Zhang, L. Liu, Y. Huang, Z. Yang, X. Lei, B. Wen, Cakdp: category- aware knowledge distillation and pruning framework for lightweight 3D object detection, in: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), 2024, pp. 15331–15341. doi:https://doi.org/10.1109/CVPR52733.2024.01452

    work page doi:10.1109/cvpr52733.2024.01452 2024

  32. [32]

    H. Song, J. Xie, Y. Duan, X. Xie, Y. Zhou, W. Wang, Cmkd-net: a cross-modal knowledge distillation method for remote sensing im- age classification, Advances in Space Research 75 (2025) 8515–8534. doi:https://doi.org/10.1016/j.asr.2024.12.XXX

    work page doi:10.1016/j.asr.2024.12.xxx 2025

  33. [33]

    H. Song, J. Xie, Liang, Y. Su, Y. Xiao, X. Zhang, Y. Ouyang, X. Li, S. Chen, Y. Li, Symmetrical learning and transferring: Efficient knowl- edge distillation for remote sensing image classification, Symmetry 17 (2025).doi:https://doi.org/10.3390/sym17010XXX

    work page doi:10.3390/sym17010xxx 2025

  34. [34]

    Umirzakova, S

    S. Umirzakova, S. Mardieva, S. Muksimova, J. Baltayev, Y. Im Cho, It- erativecontextualandadaptivestrategiesforenhancedmonoculardepth estimation, Engineering Applications of Artificial Intelligence 160 (2025) 111898.doi:https://doi.org/10.1016/j.engappai.2024.111898

    work page doi:10.1016/j.engappai.2024.111898 2025

  35. [35]

    R. Ma, Y. Zhang, B. Zhang, L. Fang, D. Huang, L. Qi, Learning at- tention in the frequency domain for flexible real photograph denois- ing, IEEE Transactions on Image Processing 33 (2024) 3707–3721. doi:https://doi.org/10.1109/tip.2024.3404253

    work page doi:10.1109/tip.2024.3404253 2024

  36. [36]

    R. Ma, S. Li, B. Zhang, L. Fang, Z. Li, Flexible and generalized real pho- tograph denoising exploiting dual meta attention, IEEE Transactions on Cybernetics 53 (10) (2023) 6395–6407.doi:https://doi.org/10. 1109/tcyb.2022.3170472

    work page arXiv 2023

  37. [37]

    R. Ma, S. Li, B. Zhang, H. Hu, Meta pid attention network for flex- ible and efficient real-world noisy image denoising, IEEE Transactions 38 on Image Processing 31 (2022) 2053–2066.doi:https://doi.org/10. 1109/tip.2022.3150294

    work page arXiv 2022

  38. [38]

    C. Miao, L. Xie, F. Wan, C. Su, H. Liu, J. Jiao, Q. Ye, Sixray: a large- scale security inspection X-ray benchmark for prohibited item discovery in overlapping images, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 2119– 2128.doi:https://doi.org/10.1109/cvpr.2019.00222

    work page doi:10.1109/cvpr.2019.00222 2019

  39. [39]

    R. Tao, Y. Wei, X. Jiang, H. Li, H. Qin, J. Wang, Y. Ma, L. Zhang, X. Liu, Towards real-world X-ray security inspection: a high-quality benchmark and lateral inhibition module for prohibited items detection (2021).doi:https://doi.org/10.1109/iccv48922.2021.01074

    work page doi:10.1109/iccv48922.2021.01074 2021

  40. [40]

    B. Wang, L. Zhang, L. Wen, X. Liu, Y. Wu, Towards real-world prohib- ited item detection: a large-scale X-ray benchmark (2021).doi:https: //doi.org/10.1109/iccv48922.2021.00536

    work page doi:10.1109/iccv48922.2021.00536 2021

  41. [41]

    S. Ren, K. He, R. Girshick, J. Sun, Faster r-cnn: towards real-time object detection with region proposal networks, in: Advances in Neural Information Processing Systems, Vol. 28, 2015, pp. 91–99.doi:https: //doi.org/10.1109/tpami.2016.2577031

    work page doi:10.1109/tpami.2016.2577031 2015

  42. [42]

    Redmon, S

    J. Redmon, S. Divvala, R. Girshick, A. Farhadi, You only look once: unified, real-time object detection, in: Proceedings of the IEEE Confer- ence on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 779–788.doi:https://doi.org/10.1109/cvpr.2016.91

    work page doi:10.1109/cvpr.2016.91 2016

  43. [43]

    YOLOv3: An Incremental Improvement

    J. Redmon, A. Farhadi, Yolov3: an incremental improvement (2018). doi:https://doi.org/10.48550/arXiv.1804.02767

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1804.02767 2018

  44. [44]

    T.-Y. Lin, P. Goyal, R. Girshick, K. He, P. Dollár, Focal loss for dense object detection, in: Proceedings of the IEEE international conference on computer vision, 2017, pp. 2980–2988.doi:https://doi.org/10. 1109/iccv.2017.324

    work page 2017

  45. [45]

    Jocher, A

    G. Jocher, A. Stoken, J. Borovec, et al., ultralytics/yolov5: v3.0, Zenodo (2020).doi:https://doi.org/10.5281/zenodo.3983579. 39

    work page doi:10.5281/zenodo.3983579 2020

  46. [46]

    Jocher, A

    G. Jocher, A. Stoken, et al., Ultralytics YOLOv8,https://github. com/ultralytics/ultralytics(2023). URLhttps://github.com/ultralytics/ultralytics

    work page 2023

  47. [47]

    Y. Zhou, X. Feng, S. Tang, J. Yang, S. Chen, X. Meng, Z. Liang, R. Ma, L. Qi, Dual attention guided context-aware feature learning for residual unfilled grains detection on threshed rice panicles, Artificial Intelligence in Agriculture 16 (1) (2026) 514–528.doi:https://doi.org/10.1016/ j.aiia.2025.11.005

    work page 2026

  48. [48]

    S. Luo, Y. Zhang, Z. Zhang, B. Guo, J. J. Lian, H. Jiang, S. Zou, W. Wang, Epdd-yolo: an efficient benchmark for pavement damage de- tection based on mamba-yolo, Measurement 253 (2025).doi:https: //doi.org/10.1016/j.measurement.2025.117638

    work page doi:10.1016/j.measurement.2025.117638 2025

  49. [49]

    Huang, Y

    X. Huang, Y. Zhang, Scanguard-YOLO: enhancing X-ray prohibited item detection with significant performance gains (2023).doi:https: //doi.org/10.3390/s24010102

    work page doi:10.3390/s24010102 2023

  50. [50]

    H. Wang, R. Tao, W. Wang, Y. Wei, Pad-f: prior-aware debias- ing framework for long-tailed X-ray prohibited item detection (2024). doi:https://doi.org/10.48550/arXiv.2411.18078

    work page doi:10.48550/arxiv.2411.18078 2024

  51. [51]

    L. Han, C. Ma, Y. Liu, J. Jia, J. Sun, Sc-YOLOv8: a security check model for the inspection of prohibited items in X-ray images, Electronics 12 (20) (2023) 4208.doi:https://doi.org/10.3390/ electronics12204208

    work page 2023

  52. [52]

    J. Cani, C. Diou, S. Evangelatos, V. Argyriou, P. Radoglou- Grammatikis, P. Sarigiannidis, I. Varlamis, G. T. Papadopoulos, Illicit objectdetectioninX-rayimagingusingdeeplearningtechniques: acom- parative evaluation (2025).doi:https://doi.org/10.48550/arXiv. 2507.17508

    work page internal anchor Pith review doi:10.48550/arxiv 2025

  53. [53]

    Zhou, K.-Y

    Y.-T. Zhou, K.-Y. Cao, D. Li, J.-C. Piao, Fine-YOLO: a simplified X- ray prohibited object detection network based on feature aggregation and normalized wasserstein distance (2024).doi:https://doi.org/ 10.3390/s24113588

    work page doi:10.3390/s24113588 2024

  54. [54]

    M. Chen, Z. Zhang, N. Jiang, X. Li, X. Zhang, YOLO-SRW: an en- hanced YOLO algorithm for detecting prohibited items in X-ray security 40 images | IEEE journals & magazine | IEEE xplore (2025).doi:https: //doi.org/10.1109/access.2025.3560840

    work page doi:10.1109/access.2025.3560840 2025

  55. [55]

    A. Haq, N. Suciati, N. D. Bui, ESI-YOLO: enhancing YOLOv8 with efficient multi-scale attention and wise-IoU for X-ray security inspec- tion, International Journal of Robotics and Control Systems 5 (3) (2025) 1770–1789.doi:https://doi.org/10.31763/ijrcs.v5i3.1983

    work page doi:10.31763/ijrcs.v5i3.1983 2025

  56. [56]

    Zhang, Y

    M. Zhang, Y. Hu, B. Xu, L. Luo, S. Wang, DSF-YOLO for weld defect detection in X-ray images with dynamic staged fusion, Scien- tific Reports 15 (1) (2025) 23305.doi:https://doi.org/10.1038/ s41598-025-06811-2

    work page 2025

  57. [57]

    Xiang, X

    Q. Xiang, X. Wang, Lei, Y. Song, Dynamic bound adaptive gradient methods with belief in observed gradients, Pattern Recognition 168 (2025).doi:https://doi.org/10.1016/j.patcog.2025.111819

    work page doi:10.1016/j.patcog.2025.111819 2025

  58. [58]

    M. Q. Ibrahim, M. Qaraad, N. K. Hussein, M. A. Farag, D. Guino- vart, Secant optimization algorithm for efficient global optimiza- tion, Scientific Reports (2026) 1–50doi:https://doi.org/10.1038/ s41598-026-36691-z

    work page 2026

  59. [59]

    N. K. Hussein, M. Qaraad, A. M. E. Najjar, M. A. Farag, M. A. El- hosseini, S. Mirjalili, D. Guinovart, Schrödinger optimizer: A quan- tum duality-driven metaheuristic for stochastic optimization and engi- neering challenges, Knowledge-Based Systems 328 (2025).doi:https: //doi.org/10.1016/j.knosys.2025.114273. 41

    work page doi:10.1016/j.knosys.2025.114273 2025