Journal of the Korean Association of Geographic Information Studies (한국지리정보학회지)

The Korean Association of Geographic Information Studies (한국지리정보학회)
권호 표지
DOI QR코드

DOI QR Code

Dataset Construction and Evaluation for Deep Learning-based Quarry Classification Using UAV Multispectral Images

UAV 다중분광 영상 기반 딥러닝 토석채취지 분류를 위한 최적 데이터세트 구성 및 평가

  • Dong-Hwan Park (Korea Forest Conservation Association) ;
  • Woo-Dam Sim (Department of Forest Management, Division of Forest Sciences, College of Forest and Environmental Sciences, Kangwon National University)
  • Received : 2024.11.25
  • Accepted : 2024.12.19
  • Published : 2024.12.30
Download PDF
⟨ Previous Next ⟩

Abstract

This study aimed to determine the optimal dataset for a deep learning-based classification model to automate and enhance the efficiency of quarry monitoring. The study site was a quarry with an area of approximately 8.5ha, located in Gumi, Gyeongsangbuk-do, where RGB and multispectral images were acquired using a DJI M3M drone. Four datasets were constructed based on the acquired images, considering spectral information composition and radiometric resolution. HRNetV2, which excels in maintaining high-resolution features, was selected as the classification model for the quarry, and the model was trained using a Hybrid Loss combining MS-SSIM, Focal Loss, and Dice Loss. The results showed that Dataset C demonstrated the most balanced classification performance with an overall accuracy of 92.8% and a Kappa coefficient of 0.875. Notably, it showed excellent performance in classifying grassland (81.0%) and wetland (87.4%), which had lower accuracy in other datasets. In contrast, Dataset D, which included DSM and Slope information, experienced overfitting problems, resulting in reduced model generalization capability. The main misclassifications occurred in the confusion between forest and grassland in areas with steep terrain changes, and grassland being misclassified as bare land in land cover boundary areas. This study examined the improvement in quarry classification accuracy according to dataset composition, including differences in spectral information configuration, radiometric resolution, and the additional use of topographic information. These findings on optimal dataset configuration are expected to serve as foundational data for the future development of automated quarry classification systems.

본 연구는 토석채취지 모니터링의 자동화 및 효율화를 위한 딥러닝 기반 분류 모델의 최적 데이터세트를 선정하고자 하였다. 연구대상지는 경상북도 구미시에 소재한 면적 약 8.5ha 규모의 토석채취 사업장으로, DJI M3M 무인항공기를 활용하여 RGB 및 멀티스펙트럴 영상을 취득하였다. 취득된 영상을 기반으로 분광정보 구성, 방사해상도를 고려하여 네 가지 데이터세트를 구성하였다. 토석채취지의 분류 모델로는 고해상도 특징 유지에 강점이 있는 HRNetV2를 선정하였으며, MS-SSIM, Focal Loss, Dice Loss를 결합한 Hybrid Loss를 활용하여 모델을 학습하였다. 연구결과, Dataset C가 전체 정확도 92.8%, Kappa 계수 0.875로 가장 균형 잡힌 분류 성능을 보였다. 특히 다른 데이터세트에서 낮은 정확도를 보인 초지(81.0%)와 습지(87.4%) 분류에서도 우수한 성능을 나타냈다. 반면, DSM과 Slope 정보를 추가한 Dataset D는 과적합 문제가 발생하여 모델의 일반화 능력이 저하되었다. 주요 오분류는 급격한 지형변화 지역의 산림지↔초지 간 혼동과 토지피복 경계부의 초지→나지 오분류로 나타났다. 본 연구는 분광정보 구성, 방사해상도에 따른 차이, 지형정보의 추가활용 등 데이터세트 구성에 따른 토석채취지 분류 정확도 향상 여부를 검토하였으며, 이러한 최적 데이터세트 구성에 대한 연구결과는 향후 토석채취지 자동 분류 시스템 개발을 위한 기초자료로 활용될 수 있을 것으로 판단된다.

Keywords

References

  1. Basu, S., S. Ganguly, S. Mukhopadhyay, R. DiBiano, M. Karki and R. Nemani. 2015. Deepsat: a learning framework for satellite imagery. In Proceedings of the 23rd SIGSPATIAL international conference on advances in geographic information systems pp. 1-10.
  2. Cheng, B., B. Xiao, J. Wang, H. Shi, T.S. Huang and L. Zhang. 2020. Higherhrnet: Scale-aware representation learning for bottom-up human pose estimation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition pp. 5386-5395.
  3. Choung. Y.J. 2014. Analysis of Land Uses in the Nakdong River Floodplain Using RapidEye Imagery and LiDAR DEM. Journal of the Korean Association of Geographic Information Studies 17(4): 189-199 (정윤재. 2021. RapidEye 영상과 LiDAR DEM을 이용한 낙동강 범람원 내 토지 이용 현황 분석. 한국지리정보학회지 17(4): 189-199). https://doi.org/10.11108/KAGIS.2014.17.4.189
  4. Delegido, J., J. Verrelst, L. Alonso and J. Moreno. 2011. Evaluation of sentinel-2 red-edge bands for empirical estimation of green LAI and chlorophyll content. Sensors 11(7): 7063-7081. https://doi.org/10.3390/s110707063
  5. FAO, F. 2010. Global forest resources assessment 2010: main report. FAO Forestry Paper 163.
  6. Garcia-Garcia, A., S. Orts-Escolano, S. Oprea, V. Villena-Martinez and J. Garcia-Rodriguez. 2017. A review on deep learning techniques applied to semantic segmentation. arXiv preprint arXiv:1704.06857.
  7. Gärtner, P., M. Förster, A. Kurban and B. Kleinschmit. 2014. Object based change detection of Central Asian Tugai vegetation with very high spatial resolution satellite imagery. International Journal of Applied Earth Observation and Geoinformation 31:110-121. https://doi.org/10.1016/j.jag.2014.03.004
  8. Huang, M.H. and R.T. Rust. 2018. Artificial intelligence in service. Journal of service research 21(2):155-172. https://doi.org/10.1177/1094670517752459
  9. Im, J., J. Rhee and J. R. Jensen. 2009. Enhancing binary change detection performance using a moving threshold window (MTW) approach. Photogrammetric Engineering & Remote Sensing 75(8): 951-961. https://doi.org/10.14358/PERS.75.8.951
  10. Kwon, S.K., E.H. Kim, J.B. Lim and A.R. Yang. 2021. The Analysis of Changes in Forest Status and Deforestation of North Korea's DMZ Using RapidEye Satellite Imagery and Google Earth. Journal of the Korean Association of Geographic Information Studies 24(4):113-126 (권수경, 김은희, 임중빈, 양아람. 2021. RapidEye 위성영상과 구글 어스를 활용한 북한 DMZ의 산림현황 및 산림황폐지 변화 분석. 한국지리정보학회지 24(4):113-126). https://doi.org/10.11108/KAGIS.2021.24.4.113
  11. LeCun, Y. 1989. Generalization and network design strategies. in Connectionism Perspective. Elsevier pp. 143–155.
  12. Li, L., Q. Yang, R. Shi and J. Teng. 2023. MSCR-HRNetV2: High-Resolution Remote Sensing Image Segmentation for Railway Scenes. In 2023 China Automation Congress (CAC) pp. 5809- 5814. IEEE.
  13. Lin, T., P. Goyal, R.B. Girshick, K. He and P. Dollár. 2017. Focal Loss for Dense Object Detection. 2017 IEEE International Conference on Computer Vision pp. 2999-3007.
  14. Liu, D. and F. Xia. 2010. Assessing object-based classification: advantages and limitations. Remote sensing letters 1(4):187-194. https://doi.org/10.1080/01431161003743173
  15. Loshchilov, I. and F. Hutter. 2017. Fixing weight decay regularization in adam. arXiv preprint arXiv: 1711.05101.
  16. Milletari, F., N. Navab and S. A. Ahmadi. 2016. V-net: Fully convolutional neural networks for volumetric medical image segmentation. In 2016 fourth internationalconference on 3D vision (3DV) pp. 565 -571. Ieee.
  17. Návar, J. 2015. Changes in Forest Production, Biomass and Carbon: Results From the 2015 UN FAO Global Forest Resource Assessment. Forest Ecology and Management 352:21-34. https://doi.org/10.1016/j.foreco.2015.05.036
  18. Park, J.H., K.D. Kim and M.J. Kang. 2016. Analysis of Quarrying and Restoration Characteristics on Quarry in Korea. Journal of Korean Society of Forest Science 105(2):223-230 (박재현, 김기대, 강민정. 2016. 국내 토석사업장의 토석채취 및 복구특성 분석. 한국산림과학회지 105(2) :223-230).
  19. Park, J.H., C.M. Park and J.W. Lee. 2010. Analysis of the Restoration Actual Conditions on the Quarries. Journal of the Korean Society of Environmental Restoration Technology 13(3):128-142 (박재현, 박종민, 이준우. 2010. 토석채취지의 복구 실태 분석. 한국환경복원기술학회지 13(3):128-142).
  20. Park, J.H. 2012. Analysis of the Restoration Actual Conditions on the Quarries II. Journal of Korean Forest Society 101(4):558-566 (박재현. 2012. 토석채취지의 복구 실태 분석 Ⅱ. 한국산림과학회지 101(4):558-566).
  21. Park. J.H. and T.R. Kim. 2021. Analysis of Drone Surveying Using a Low-Cost PPK Kit. Journal of the korean gromophological association 28(4):41-52 (박준호, 김태림. 2021. PPK Kit를 활용한 드론 측량분석. 한국지형학회지 28(4): 41-52).
  22. Rouhi, R., M. Jafari, S. Kasaei and P. Keshavarzian. 2015. Benign and malignant breast tumors classification based on region growing and CNN segmentation. Expert Systems with Applications 42(3): 990-1002. https://doi.org/10.1016/j.eswa.2014.09.020
  23. Rouse, J.W., R.H. Haas, J.A. Schell and D.W. Deering. 1974. Monitoring vegetation systems in the Great Plains with ERTS. NASA Spec. Publ 351(1): 309.
  24. Schöpfer, E., S. Lang and J. Strobl. 2010. Segmentation and object-based image analysis. Remote sensing of urban and suburban areas 181-192.
  25. Smith, L.N. and N. Topin. 2019. Superconvergence: Very fast training of neural networks using large learning rates. In Artificial intelligence and machine learning for multi-domain operations applications. pp. 369-386. SPIE.
  26. Sun, K., B. Xiao, D. Liu and J. Wang. 2019. Deep high-resolution representation learning for human pose estimation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition pp. 5693-5703.
  27. Wang, J., K. Sun, T. Cheng, B. Jiang, C. Deng, Y. Zhao, D. Liu, Y. Mu, M. Tan, X. Wang, W. Liu and B. Xiao. 2020. Deep high-resolution representation learning for visual recognition. IEEE transactions on pattern analysis and machine intelligence 43(10):3349-3364. https://doi.org/10.1109/TPAMI.2020.2983686
  28. Wang, Z., E.P. Simoncelli and A.C. Bovik. 2003. Multiscale structural similarity for image quality assessment. In The Thrity-Seventh Asilomar Conference on Signals, Systems & Computers pp. 1398 -1402.
  29. Won, M.S., Y.S. Kim and K.H. Kim. 2014. Estimation on Greenhouse Gases(GHGs) Emission of Large Forest Fire Area in 2013. Journal of the Korean Association of Geographic Information Studies 17(3): 54-67 (원명수, 김유승, 김경하. 2014. RapidEye 영상을 활용한 대형산불피해지의 온실가스 배출량 추정. 한국지리정보학회지 17(3):54-67). https://doi.org/10.11108/KAGIS.2014.17.3.054