Skip to main content
SpringerLink
Log in

EBL: Efficient background learning for x-ray security inspection

  • Published:
Applied Intelligence Aims and scope Submit manuscript

Abstract

A significant issue in X-ray security inspection is that there are considerably more background images than foreground images. Although convolutional neural network (CNN)-based detection models have proven to be effective at detecting objects in X-ray security inspection systems, they are not designed to handle imbalance issues during training. Hence, typical CNN models are suboptimal at extracting background features and balancing the foreground and background. To address these two problems, we propose an efficient background learning (EBL) method with three modules: mixed foreground and background learning (MFB), hierarchical balanced hard negative example (HBHE) sampler and prime background mining with voting (PBMV). The MFB module extracts the foreground and background during each iteration and combines them into a single image for training, achieving unified and balanced foreground and background training. The HBHE sampler balances difficult foreground and background samples, dynamically choosing the number of negative samples based on the MFB image by calculating the difficulty factor. The PBMV module selects a prime background that is prone to false detection and provides increased attention through multiple voting during training. Experiments show that EBL can reduce false positives while maintaining high recall. When applied to Faster R-CNN, AP50 increases by 5.8% on benchmark X-ray datasets, including 2.3% on OPIXray and 3.7% on SIXray. Moreover, the performance of our model is improved without increasing computational costs or memory during the inference phase.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8
Fig. 9
Fig. 10

Similar content being viewed by others

Data Availability

The datasets analysed during the current study are available at:

• OPIXray: https://github.com/OPIXray-author/OPIXray

• SIXray: https://github.com/MeioJane/SIXray

• EDXray: not available due to privacy policy.

References

  1. Miao C, Xie L, Wan f et al (2019) 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, pp 2119–2128

  2. Mery D, Riffo V, Zscherpel U et al (2015) GDXRay: The database of X-ray images for nondestructive testing. J Nondestruct Eval 34(4):1–12

    Article  Google Scholar 

  3. Wei Y, Tao R, Wu Z et al (2020) Occluded prohibited items detection: An x-ray security inspection benchmark and de-occlusion attention module. In: Proceedings of the 28th ACM international conference on multimedia, pp 138–146

  4. Wang B, Zhang L, Wen L et al (2021) Towards real-world prohibited item detection: A large-scale x-ray benchmark. In: Proceedings of the IEEE/CVF international conference on computer vision, pp 5412–5421

  5. Girshick R, Donahue J, Darrell T et al (2014) Rich feature hierarchies for accurate object detection and semantic segmentation. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 580–587

  6. Girshick R (2015) Fast r-cnn. In: Proceedings of the IEEE international conference on computer vision, pp 1440–1448

  7. Ren S, He K, Girshick R et al (2015) Faster r-cnn: Towards real-time object detection with region proposal networks. Advances in neural information processing systems, vol 28

  8. Cai Z (2018) Vasconcelos N. Cascade r-cnn: Delving into high quality object detection. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 6154–6162

  9. Lin TY, Dollár P, Girshick R et al (2017) Feature pyramid networks for object detection. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 2117–2125

  10. Lin TY, Goyal P, Girshick R et al (2017) Focal loss for dense object detection. In: Proceedings of the IEEE international conference on computer vision, pp 2980–2988

  11. Liu W, Anguelov D, Erhan D et al (2016) Ssd: Single shot multibox detector. In: European conference on computer vision. Springer, Cham, pp 21–37

  12. Redmon J, Divvala S, Girshick R et al (2016) You only look once: Unified, real-time object detection. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 779–788

  13. Redmon J, Farhadi A (2017) YOLO9000: better, faster, stronger. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 7263–7271

  14. Redmon J, Farhadi A (2018) Yolov3: An incre-mental improvement. arXiv:1804.02767

  15. Bochkovskiy A, Wang CY, Liao HYM (2020) Yolov4: Optimal speed and accuracy of object detection. arXiv:2004.10934

  16. Wang CY, Bochkovskiy A, Liao HYM (2021) Scaled-yolov4: Scaling cross stage partial network. In: Proceedings of the IEEE/cvf conference on computer vision and pattern recognition, pp 13029–13038

  17. Tian Z, Shen C, Chen H et al (2019) Fcos: Fully convolutional one-stage object detection. In: Proceedings of the IEEE/CVF international conference on computer vision, pp 9627–9636

  18. Duan K, Bai S, Xie L et al (2019) Centernet: Keypoint triplets for object detection. In: Proceedings of the IEEE/CVF international conference on computer vision, pp 6569–6578

  19. Yun S, Han D, Oh SJ et al (2019) Cutmix: Regularization strategy to train strong classifiers with localizable features. In: Proceedings of the IEEE/CVF international conference on computer vision, pp 6023–6032

  20. Shrivastava A, Gupta A, Girshick R (2016) Training region-based object detectors with online hard example mining. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 761–769

  21. Cao Y, Chen K, Loy CC et al (2020) Prime sample attention in object detection. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp 11583–11591

  22. Bradley D, Roth G (2007) Adaptive thresholding using the integral image. J Graph Tools 12 (2):13–21

    Article  Google Scholar 

  23. He K, Zhang X, Ren S et al (2016) Deep residual learning for image recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 770–778

  24. Chen K, Wang J, Pang J et al (2019) MMDetection: Open mmlab detection toolbox and benchmark. arXiv:1906.07155

  25. Lin TY, Maire M, Belongie S et al (2014) Microsoft coco: Common objects in context. In: European conference on computer vision. Springer, Cham, pp 740–755

  26. LeCun Y, Bengio Y, Hinton G (2015) Deep learning. nature 521(7553):436–444

    Article  Google Scholar 

  27. Akcay S, Breckon TP (2017) An evaluation of region based object detection strategies within x-ray baggage security imagery. In: 2017 IEEE International conference on image processing (ICIP). IEEE, pp 1337–1341

  28. Akçay S, Kundegorski ME, Devereux M et al (2016) Transfer learning using convolutional neural networks for object classification within X-ray baggage security imagery. In: 2016 IEEE International conference on image processing (ICIP). IEEE, pp 1057–1061

  29. Pang J, Chen K, Shi J et al (2019) Libra r-cnn: Towards balanced learning for object detection. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp 821–830

  30. Zhang H, Chang H, Ma B et al (2020) Dynamic r-CNN: Towards high quality object detection via dynamic training. In: European conference on computer vision. Springer, Cham, pp 260–275

  31. Law H, Deng J (2018) Cornernet: Detecting objects as paired keypoints. In: Proceedings of the European conference on computer vision (ECCV), pp 734–750

  32. Zhang H, Cisse M, Dauphin YN et al (2017) Mixup: Beyond empirical risk minimization. arXiv:1710.09412,

  33. Gao SH, Cheng MM, Zhao K et al (2019) Res2net: A new multi-scale backbone architecture. IEEE Trans Pattern Anal Mach Intell 43(2):652–662

    Article  Google Scholar 

  34. Xie S, Girshick R, Dollár P et al (2017) Aggregated residual transformations for deep neural networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 1492–1500

Download references

Acknowledgements

This work was supported by the National Key Research and Development Program of China(SQ2020YFF0426385). The authors would like to thank AJE (www.aje.com) for its linguistic assistance during the preparation of this manuscript.

Author information

Authors and Affiliations

  1. Zhejiang PeckerAI Technology., Ltd, Binjiang, Hangzhou, 310000, Zhejiang, China

    Wei Wang, Linyang He, Yiqing Li, Kai Zhou, Linchao Li, Guohua Cheng & Ting Wen

Authors
  1. Wei Wang
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Linyang He
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Yiqing Li
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Kai Zhou
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Linchao Li
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Guohua Cheng
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. Ting Wen
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Yiqing Li.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendices

Appendix A: Effect of the Hyperparameters

We conduct comparative experiments on some hyperparameters in the HBHE and PBMV modules. The experimental results for different δ and k are shown in Table 9. Smaller values of δ address the problem of extreme sampling bias to a certain extent. We find that setting δ to 0.1 achieves the best trade-off and use this as the default value in later experiments. Larger values of k reduce the overall quality of the difficult backgrounds. We find that setting k to 0.1 achieves the best performance. The experimental results on the resampling times are shown in Table 10. We find that a prime background that appears only once can affect the performance of the model if that background participates in resampling.

Table 9 Ablation study on δ and k
Full size table
Table 10 Ablation study on α, β and γ
Full size table

Appendix B: Result visualization

We visualize the background mining results of PBMV, as shown in Fig. 11. Nonprime BGs have relatively simple features that are difficult to falsely detect and are ineffective for training the model. The “Prime” background usually contains many objects with more complex features. The “S-Prime” background contains even more complex features than the Prime background. The features in these backgrounds are useful for reducing the number of false positives in the model.

Fig. 11
figure 11

Visualization of the PBMV results. “N-Prime” represents the nonprime BGs. “Prime” represents the prime BGs of the mining results, which appear only once during hard voting. “S-Prime” represents super prime BGs, which appear many times during prime voting

Full size image

Figure 12 visualizes the detection results of some examples and compares the baseline with our method on both the FGs and BGs. In the FGs, our method not only ensurethat contraband is detected but also eliminate falsely detected objects in the background that are close to the baseline. In the BGs, our method greatly reduces false detection.

Fig. 12
figure 12

Visualization of the detection results. The red boxes represent the ground truth boxes, the blue boxes represent the baseline results and the green boxes represent the results of our method

Full size image

Rights and permissions

Springer Nature or its licensor holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Wang, W., He, L., Li, Y. et al. EBL: Efficient background learning for x-ray security inspection. Appl Intell 53, 11357–11372 (2023). https://doi.org/10.1007/s10489-022-04075-1

Download citation

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10489-022-04075-1

Keywords

Search

Navigation