|
Abstract.
The introduction of automated detection technologies using artificial intelligence to process the results of non-intrusive control over the movement of prohibited goods will increase public confidence in the technologies used to protect against destructive actions in places where large numbers of people gather during mass sports, cultural or educational events. The article presents a methodology for constructing ensemble models to solve recognition and decision automation tasks involving images from non-intrusive product inspection systems. The detection of prohibited items is considered as a practical example. Object detection neural networks are based on the YOLOv8x architecture, and EfficientNet-B6 is used for image classification neural networks. The performance of the models is assessed using accuracy, repeatability, and average accuracy (AP) metrics for both individual detection networks and the ensemble. The results obtained highlight the advantages and limitations of the proposed approach. The methodology can also be extended to other tasks of recognition and automation of decision-making in X-ray inspection of goods and vehicles, such as determining specific categories of goods, nomenclatures or their groupings.
Keywords:
images from cargo and vehicle X-ray inspection systems, ensemble modeling, YOLOv8x, EfficientNet-B6, improvement of detection accuracy, neural networks, deep learning.
DOI 10.14357/20718632260213
EDN PNTUON
PP. 129-141.
References
1. Bulavin V.I., Skiba V.Yu. et al.: 30 years of automation of customs authorities. From the idea of creating automated information processing systems in state customs control bodies to the Strategy for the Development of the Customs Service of the Russian Federation until 2030. Moscow: Central Base Customs House, 2021. 304 p. (In Russ.). EDN: SAYHWT. 2. Skiba V.Yu., Pozdnyakova K.E. Modern automated information systems for customs operations without the participation of customs officials. Vestnic of Russian Customs Academy. 2022; (2 (59)): 19 - 33. (In Russ.). EDN: UCKJDF. https://doi.org/10.54048/20727240_2022_02_19 3. Skiba V.Yu., Bykov A.M. Implementation of digital technologies in the institution of customs payments. Vestnic of the Russian Customs Academy. 2022; (2 (59)): 42 - 51. (In Russ.). EDN GURZIQ. https://doi.org/10.54048/20727240_2022_02_42 4. Fuchs E., Novikov V., Novikov S. Improvements in post-release customs control of goods moved by different modes of transport. Transportation Research Procedia. 2022. (61): 200 - 205. https://doi.org/10.1016/j.tpro.2022.01.033. 5. Customs inspection with ISC. Available from: https://avangard-direct.ru/blog/tamozhennyj-inspekcionno-dosmotrovyj-kompleks/ [Accessed 25 March 2026]. 6. Bashly P.N., Verbov V.F. Application of machine learning for automatic analysis of images of inspection and screening systems. Vestnic of the Russian Customs Academy. 2022; (2 (59)): 123 - 132. (In Russ.). EDN MMNGZZ. https://doi.org/10.54048/20727240_2022_02_123 7. Krasnova A.I., Pulin A.O. Analysis of customs control objects recognized using ISC. Bulletin of innovative technologies. 2019; 3 (4 (12)): 40 - 43. (In Russ.). EDN JZNLVA. 8. Denis Tereshchenko spoke about the model of an intelligent checkpoint being developed by the Federal Customs Service of Russia at the Eurasian Economic Forum 2022. Available from: https://customs.gov.ru/press/federal/document/341465 [Accessed 25 March 2026]. 9. Dumoulin V., Visin F.A. A guide to convolution arithmetic for deep learning. arXiv preprint arXiv:1603.07285. https://arxiv.org/pdf/1603.07285. pdf [Accessed 25 March 2026]. 10. Dhiraj N., Jain D.K. An evaluation of deep learning based object detection strategies for threat object detection in baggage security imagery. Pattern Recognition Letters. 2019; (120): 112 - 119. https://doi.org/10.1016/j.patrec.2019.01.014 11. Unified Commodity Nomenclature of Foreign Economic Activity of the Eurasian Economic Union and the Unified Customs Tariff of the Eurasian Economic Union. Available from: https://ecc.eaeunion.org/commission/department/catr/et/ [Accessed 25 March 2026]. 12. Zhang X., Razavi-Far R., Isah H., David A., Higgins G., Lu R., Ghorbani A.A. Area in circle: A novel evaluation metric for object detection. Knowledge-Based Systems. 2024; (293): 111684. https://doi.org/10.1016/j.knosys.2024.111684 13. Yang J., Liu Z. A novel real-time steel surface defect detection method with enhanced feature extraction and adaptive fusion. Engineering Applications of Artificial Intelligence. 2024; (138): 109289. https://doi.org/10.1016/j.engappai.2024.109289 14. Wang X., Tian Y., Zheng K., Liu C. C2Net-YOLOv5: A bidirectional Res2Net-based traffic sign detection algorithm. Computers, Materials & Continua. 2023. 77 (2): 1949 - 1965. https://doi.org/10.32604/cmc.2023.042224 15. Saakyan A. Cascade models in computer vision to boost accuracy and performance. Available from: https://www.conf42.com/Machine_Learning_2023_Argo_Saakyan_cascade _models_computer_vision_accuracy_performance [Accessed 25 March 2026]. 16. Hamdi E.B., Hidayaturrahman: Ensemble of pre-trained vision transformer models in plant disease classification, an efficient approach. Procedia Computer Science. 2024. (245): 565 - 573. https://doi.org/10.1016/j.procs.2024.10.283 17. Park J., Lee J., Jeong J. YOLOv5 based object detection in reel package X-ray images of semiconductor component. Heliyon. 2024. 10 (5): e26532. https://doi.org/10.1016/j.heliyon.2024.e26532 18. Explore Ultralytics YOLOv8. Available from: https://docs.ultralytics.com/ru/models/yolov8/ [Accessed 25 March 2026]. 19. Vasanthi P., Laavanya M. Efficient YOLOv8 algorithm for extreme small-scale object detection. Digital Signal Processing. 2024. (145): 104682. https://doi.org/10.1016/j.dsp.2024.104682 20. Ultralytics YOLO11. Available from: https://docs.ultralytics.com/ru/models/yolo11/ [Accessed 25 March 2026]. 21. Sharma A., Kumar V., Longchamps L. Comparative performance of YOLOv8, YOLOv9, YOLOv10, YOLOv11 and Faster R-CNN models for detection of multiple weed species. Smart Agricultural Technology. 2024. (9): 100648. https://doi.org/10.1016/j.atech.2024.100648 22. Tan M., Le Q. Efficientnet: Rethinking model scaling for convolutional neural networks. In: International Conference on Machine Learning. PMLR. 2019. P. 6105 - 6114. 23. Yadav P., Menon N., Ravi V., Vishvanathan S., Pham T.D. EfficientNet convolutional neural networks-based Android malware detection. Computers & Security. 2022. (115): 102622. https://doi.org/10.1016/j.cose.2022.102622 24. Tan M., Le Q.V. Efficientnetv2: Smaller models and faster training. In: International conference on machine learning. PMLR. 2021. P. 10096 - 10106. 25. Zhao Z., Bakar E.B.A., Razak N.B.A., Akhtar M.N. Corrosion image classification method based on EfficientNetV2. Heliyon. 2024. 10(17): e36754. https://doi.org/10.1016/j.heliyon.2024.e36754 26. Kang T., Wu Y. Rethinking PASCAL-VOC and MS-COCO dataset for small object detection. Journal of Visual Communication and Image Representation. 2023. (93): 103830. https://doi.org/10.1016/j.jvcir.2023.103830
|
