AccScience Publishing / IJOCTA / Online First / DOI: 10.36922/IJOCTA026170066
Submit to IJOCTA
Cite this article
2
Download
44
Views
Related Info Links
Google Scholar
More by Authors Links
Jih-Gau Juang
Journal Browser
Volume | Year
Issue
Search
Forthcoming Issue
Current Issue
View All
News and Announcements
View All
RESEARCH ARTICLE

Optimization of aerial image stitching for garbage identification

Cheng-Chien Su1 ,  Jih-Gau Juang2*
Show Less
1 Department of Hardware R&D, Wincomm Corporation, Hsinchu, Taiwan
2 Department of Communications, Navigation and Control Engineering, Faculty of Electrical Engineering and Computer Science, National Taiwan Ocean University, Keelung, Taiwan
IJOCTA, 026170066 https://doi.org/10.36922/IJOCTA026170066
Received: 23 April 2026 | Revised: 22 June 2026 | Accepted: 23 June 2026 | Published online: 8 July 2026
© 2026 by the Author(s). This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution -Noncommercial 4.0 International License (CC-by the license) ( https://creativecommons.org/licenses/by-nc/4.0/ )
Download PDF
XML
Cite
Abstract

To improve the environmental quality and protect local ecosystems, riverine waste monitoring systems are needed to identify waste along riverbanks. Since most riverbank areas are wide, a single unmanned aerial vehicle (UAV) camera cannot capture the entire area. Hence, the use of multiple UAVs and image stitching for river scene reconstruction is required. This study applies image processing with multiple UAVs to identify garbage. First, image stitching employed the scale-invariant feature transform, Oriented Features from Accelerated Segment Test and Rotated Binary Robust Independent Elementary Features, and accelerated KAZE for feature detection and description. Real-time scene stitching was performed using feature point matching and filtering, and seam removal was achieved with an averaging-weighted mask. Two image-stitching datasets were tested, and the proposed method preserves 100% of the content of the original images. The computation time was reduced by 60% compared to the conventional method. Secondly, the optimized stitching results were used to further identify marine pollution using the You Only Look Once (YOLO) model.  Different types of YOLO networks were used to identify waste pollution. An optimal YOLO model that identified waste items with 100% recognition and accuracy was obtained. The current coordinates, waste types, area size, and real-time images of the waste pollution were sent to the monitoring terminal. This enabled the analysis of waste distribution and facilitated attempts to trace the source of the waste by comparing training results and the identification effectiveness of different models.

Keywords
Unmanned aerial vehicles
Image identification
Dynamic image stitching
Deep learning
Stitch fusion
Funding
National Science and Technology Council, Taiwan (NSTC 112-2221-E-019 -040-MY2).
Conflict of interest
The authors declare they have no competing interests.
References
  1. Liao YH, Juang JG. Automatic marine debris inspection. Aerospace. 2023;10(1):84. https://doi.org/10.3390/aerospace10010084
  2. Liao S. HAIDA-trash-dataset-high-resolution-aerial-image. Published 2021. Accessed April 25, 2025. https://github.com/LiaoSteve/HAIDA-Trash-Dataset-High-Resolution-Aerial-image
  3. He X, He L, Li X. Image stitching via convolutional neural network. In: 2021 7th International Conference on Computer and Communications (ICCC). Piscataway, NJ: IEEE; 2021:709-713. https://doi.org/10.1109/ICCC54389.2021.9674411
  4. Kumareswaran D, Nasir N, Rahman MZA, et al. Computational speed and qualitative assessment of real-time image stitching algorithm. In: 2021 International Conference on Communication, Control and Information Sciences (ICCISc). Piscataway, NJ: IEEE; 2021:1-6. https://doi.org/10.1109/ICCISc52257.2021.9484887
  5. Lowe DG. Distinctive image features from scale-invariant keypoints. Int J Comput Vis. 2004;60(2):91-110. https://doi.org/10.1023/B:VISI.0000029664.99615.94
  6. Rublee E, Rabaud V, Konolige K, et al. ORB: an efficient alternative to SIFT or SURF. In: 2011 International Conference on Computer Vision. Piscataway, NJ: IEEE; 2011:2564-2571. https://doi.org/10.1109/ICCV.2011.6126544
  7. Alcantarilla P, Nuevo J, Bartoli A. Fast explicit diffusion for accelerated features in nonlinear scale spaces. In: Proceedings of the British Machine Vision Conference 2013. Durham, UK: British Machine Vision Association; 2013:13.1-13.11. https://doi.org/10.5244/C.27.13
  8. Jakubović S, Velagic J. Image feature matching and object detection using brute-force matchers. In: 2018 International Symposium ELMAR. Piscataway, NJ: IEEE; 2018:83-86. https://doi.org/10.23919/ELMAR.2018.8534641
  9. OpenCV. Basics of brute-force matcher. Accessed June 15, 2025. https://docs.opencv.org/4.x/dc/dc3/tutorial_py_matcher.html
  10. Yang Z, Qin X. Image stitching technology based on SIFT, FLANN, and RPOSAC algorithms. In: 2023 4th International Conference on Computer Vision, Image and Deep Learning (CVIDL). Piscataway, NJ: IEEE; 2023:52-56. https://doi.org/10.1109/CVIDL58838.2023.10167270
  11. Muja M, Lowe DG. FLANN-fast library for approximate nearest neighbors. In: Proceedings of the Fourth International Conference on Computer Vision Theory and Applications. Setúbal, Portugal: SciTePress - Science and Technology Publications; 2009;2:331-340. Accessed April 20, 2025. https://www.scienceopen.com/document?vid=a6407082-f90a-4b9b-aa76-0b7796497bcd
  12. Zhang Y, Zhang J, Zhang L, et al. Research on panorama reconstruction technique of UAV aerial image based on improved ORB algorithm. In: 2019 3rd International Conference on Electronic Information Technology and Computer Engineering (EITCE). Piscataway, NJ: IEEE; 2019:1252-1256. https://doi.org/10.1109/EITCE47263.2019.9095113
  13. OpenCV. Understanding k-nearest neighbors. Accessed January 11, 2025. https://docs.opencv.org/4.x/d5/d26/tutorial_py_knn_understanding.html
  14. Fischler MA, Bolles RC. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography. In: Readings in Computer Vision. Amsterdam, Netherlands: Elsevier; 1987:726-740. https://doi.org/10.1016/B978-0-08-051581-6.50070-2
  15. Gan W, Wu Z, Wang M, Cui X. Image stitching based on optimized SIFT algorithm. In: Proceedings of 2023 5th International Conference on Intelligent Control, Measurement and Signal Processing (ICMSP). Piscataway, NJ: IEEE; 2023:1099-1102. https://doi.org/10.1109/ICMSP58539.2023.10170989
  16. Chum O, Matas J. Matching with PROSAC - progressive sample consensus. In: 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05). Piscataway, NJ: IEEE; 2005;1:220-226. https://doi.org/10.1109/CVPR.2005.221
  17. Jeon HK, Jeong JM, Lee KY. An implementation of the real-time panoramic image stitching using ORB and PROSAC. In: 2015 International SoC Design Conference (ISOCC). Piscataway, NJ: IEEE; 2015:91-92. https://doi.org/10.1109/ISOCC.2015.7401661
  18. Li H, Wang J, Han C. Image mosaic and hybrid fusion algorithm based on pyramid decomposition. In: 2020 International Conference on Virtual Reality and Visualization (ICVRV). Piscataway, NJ: IEEE; 2020:205-208. https://doi.org/10.1109/ICVRV51359.2020.00049
  19. Xiu C, Ma Y. Image stitching based on improved gradual fusion algorithm. In: 2019 Chinese Control and Decision Conference (CCDC). Piscataway, NJ: IEEE; 2019:2933-2937. https://doi.org/10.1109/CCDC.2019.8832814
  20. Wang SI. Title in Pinyin Chinese [Yingyong Dichengben Shuangpin GNSS RTK Jishu Yu Wurenji Dingwei Dingxiang Zhi Yanjiu]. J Photogramm Remote Sens. 2020;25(4):241-252. [In Chinese]. https://doi.org/10.6574/JPRS.202012_25(4).0004
  21. Jocher G, Chaurasia A, Qiu J. Ultralytics YOLO (Version 8.0.0) [computer software]. Accessed August 27, 2024. https://github.com/ultralytics/ultralytics
  22. Bawankule R, Gaikwad V, Kulkarni I, et al. Visual detection of waste using YOLOv8. In: 2023 International Conference on Sustainable Computing and Smart Systems (ICSCSS). Piscataway, NJ: IEEE; 2023:869-873. https://doi.org/10.1109/ICSCSS57650.2023.10169688
  23. Sapkota R, Ahmed D, Karkee M. Comparing YOLOv8 and Mask R-CNN for instance segmentation in complex orchard environments. Artif Intell Agric. 2024;13:84-99. https://doi.org/10.1016/j.aiia.2024.07.001
  24. Logitech BRIO ULTRA HD PRO webcam. Accessed August 2, 2024. https://www.logitech.com/zh-tw/products/webcams/brio-4k-hdr-webcam.960-001105.html
  25. Bass J. imagezmq: A Set of Python Classes That Transport OpenCV Images from One Computer to Another Using PyZMQ Messaging. Accessed March 12, 2025. https://github.com/jeffbass/imagezmq#why-use-imagezmq
  26. Gimbal Overview. ArduPilot. Accessed March 2, 2025. https://ardupilot.org/copter/docs/common-tarot-gimbal.html
  27. Rosten E, Drummond T. Machine learning for high-speed corner detection. In: Lecture Notes in Computer Science. Berlin, Germany: Springer; 2006:430-443. https://doi.org/10.1007/11744023_34
  28. Calonder M, Lepetit V, Strecha C, et al. BRIEF: binary robust independent elementary features. In: Lecture Notes in Computer Science. Berlin, Germany: Springer; 2010:778-792. https://doi.org/10.1007/978-3-642-15561-1_56
  29. Homography matrix transformation. Accessed September 12, 2024. https://lxrd-aj.github.io/blog/cv/homography/
  30. Chiang CH, Juang JG. Application of UAVs and image processing for riverbank inspection. Machines. 2023;11(9):876. https://doi.org/10.3390/machines11090876
  31. Ultralytics YOLOv5 architecture. Accessed March 2, 2025. https://docs.ultralytics.com/yolov5/tutorials/architecture_description/#41-compute-losses
  32. YOLOv5 architecture description. arXiv. 2024. Accessed March 2, 2025. https://arxiv.org/html/2407.20892v1
  33. What is YOLOv8? Roboflow Blog. Accessed May 2, 2025. https://blog.roboflow.com/whats-new-in-yolov8/
  34. YOLOv8 architecture description. Medium. Accessed May 2, 2025. https://henry870603.medium.com/object-detection-yolov8%E8%A9%B3%E8%A7%A3-fdf8874e5e99
  35. About the functionality of DFL_loss. GitHub Issues. Accessed May 2, 2025. https://github.com/ultralytics/ultralytics/issues/4219?trk=article-ssr-frontend-pulse_little-text-block
  36. YOLO8 loss function for classification task. GitHub Issues. Accessed May 2, 2025. https://github.com/ultralytics/ultralytics/issues/4684?trk=article-ssr-frontend-pulse_little-text-block
  37. Feng C, Zhong Y, Gao Y, et al. TOOD: task-aligned one-stage object detection. In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway, NJ: IEEE; 2021:3490-3499. https://doi.org/10.1109/ICCV48922.2021.00349
  38. What is instance segmentation? Roboflow Blog. Accessed October 23, 2025. https://blog.roboflow.com/instance-segmentation-roboflow/
  39. Perera CJ, Premachandra C, Kawanaka H. Low pixel resolution hyperspectral image mosaics generation using learning-based feature matching. IEEE Access. 2023;11:104084-104093. https://doi.org/10.1109/ACCESS.2023.3315769
  40. Liu S, Chai Q. Shape-optimizing and illumination-smoothing image stitching. IEEE Trans Multimedia. 2019;21(3):690-703. https://doi.org/10.1109/TMM.2018.2864576
  41. Nakaoka Y, Asaki S, Premachandra C. E-YOLO to OMOD: an enhanced YOLO framework for small moving object detection in omnidirectional videos. IEEE Open J Intell Trans Syst. 2026;7:838-846. https://doi.org/10.1109/OJITS.2026.3670217
  42. Citroni R, Di Paolo F, Livreri P. A novel energy harvester for powering small UAVs: performance analysis, model validation and flight results. Sensors. 2019;19(8):1771. https://doi.org/10.3390/s19081771
Next article in this issue
Share
Back to top
An International Journal of Optimization and Control: Theories & Applications, Electronic ISSN: 2146-5703 Print ISSN: 2146-0957, Published by AccScience Publishing
We use cookies on our website to ensure you get the best experience.
Read more about our cookies here
Accept