DOI QR코드

DOI QR Code

Selection of Optimal Band Combination for Machine Learning-based Water Body Extraction using SAR Satellite Images

SAR 위성 영상을 이용한 수계탐지의 최적 머신러닝 밴드 조합 연구

  • Jeon, Hyungyun (School of Earth and Environmental Sciences, Seoul National University) ;
  • Kim, Duk-jin (School of Earth and Environmental Sciences, Seoul National University) ;
  • Kim, Junwoo (School of Earth and Environmental Sciences, Seoul National University) ;
  • Vadivel, Suresh Krishnan Palanisamy (School of Earth and Environmental Sciences, Seoul National University) ;
  • Kim, JaeEon (School of Earth and Environmental Sciences, Seoul National University) ;
  • Kim, Taecin (School of Earth and Environmental Sciences, Seoul National University) ;
  • Jeong, SeungHwan (School of Earth and Environmental Sciences, Seoul National University)
  • Received : 2020.07.29
  • Accepted : 2020.08.24
  • Published : 2020.09.30

Abstract

Water body detection using remote sensing based on machine interpretation of satellite image is efficient for managing water resource, drought and flood monitoring. In this study, water body detection with SAR satellite image based on machine learning was performed. However, non water body area can be misclassified to water body because of shadow effect or objects that have similar scattering characteristic comparing to water body, such as roads. To decrease misclassifying, 8 combination of morphology open filtered band, DEM band, curvature band and Cosmo-SkyMed SAR satellite image band about Mokpo region were trained to semantic segmentation machine learning models, respectively. For 8 case of machine learning models, global accuracy that is final test result was computed. Furthermore, concordance rate between landcover data of Mokpo region was calculated. In conclusion, combination of SAR satellite image, morphology open filtered band, DEM band and curvature band showed best result in global accuracy and concordance rate with landcover data. In that case, global accuracy was 95.07% and concordance rate with landcover data was 89.93%.

인공위성 영상을 기반으로 한 기계판독(machine interpretation) 원격탐사 수계 탐지는 효율적인 수자원 관리, 가뭄 탐지, 홍수 모니터링 등에 큰 도움이 된다. 따라서 본 연구에서는 머신러닝을 기반으로 한 SAR 위성 영상 기반 수계 탐지를 시행하였다. 그러나 SAR 위성 영상만을 사용하였을 경우 음영 효과 또는 도로 등의 수계와 비슷한 산란특성을 가지는 물체로 인하여 비수계가 수계로 오탐지 될 수 있다. 이러한 오탐지를 줄이기 위하여 목포 지역을 촬영한 Cosmo-SkyMed SAR 위성 영상에 모폴로지(Morphology)의 open 연산을 거친 밴드와 DEM(수치표고모델) 밴드, Curvature(곡률) 밴드를 조합하여 중첩한 8가지 경우에 대하여 의미 분할 기법 머신러닝 모델을 학습시켰다. 8가지 머신러닝 모델에 대한 최종 테스트 결과인 Global Accuracy를 구하였으며, 목포 지역의 토지피복지도와의 일치율 역시 비교하였다. 그 결과 SAR 위성 영상과 모폴로지 open 필터를 적용한 밴드, DEM 밴드, Curvature 밴드를 모두 사용한 경우가 Global Accuracy뿐만 아니라 토지피복지도와의 일치율 역시 가장 높음을 확인할 수 있었다. 이때 Global Accuracy는 95.07%였으며, 토지피복지도와의 일치율은 89.93%로 나타났다.

Keywords

References

  1. Badrinarayanan, V., A. Kendall and R. Cipolla. 2017. Segnet: A deep convolutional encoder-decoder architecture for image segmentation. IEEE Transactions on pattern analysis and machine intelligence 39(12):2481-2495. https://doi.org/10.1109/TPAMI.2016.2644615
  2. Dou, J., H. Yamagishi, H. R. Pourghasemi, A. P. Yunus, X. Song, Y. Xu and Z. Zhu. 2015. An integrated artificial neural network model for the landslide susceptibility assessment of Osado Island. Japan Natural Hazards 78(3):1749-1776. https://doi.org/10.1007/s11069-015-1799-2
  3. Hong, S., H. Jang, N. Kim and H. Sohn. 2015. Water area extraction using RADARSAT SAR imagery combined with landsat imagery and terrain information. Sensors 15(3):6652-6667. https://doi.org/10.3390/s150306652
  4. Huang, W., B. DeVries, C. Huang, M. W. Lang, J. W. Jones, I. F. Creed and M. L. Carroll. 2018. Automated extraction of surface water extent from Sentinel-1 data. Remote Sensing 10(5):797. https://doi.org/10.3390/rs10050797
  5. Hwang, S. K. 2019. Computer vision and machine learning with Opencv4. Gilbut, Seoul, pp.369-378..
  6. Kim, Y., M. Lee and S. Lee. 2019. Detection of change in water system due to collapse of Laos Xe pian-Xe namnoy dam using KOMPSAT-5 satellites, Korean Journal of Remote Sensing 35(6_4):1417-1424. https://doi.org/10.7780/kjrs.2019.35.6.4.10
  7. Lee, S. B., Y. Kim, J. Kim and Y. Park. 2017. Detection of alteration in river area using Landsat satellite imagery. Journal of the Korean Society of Hazard Mitigation 17(3):391-398. https://doi.org/10.9798/KOSHAM.2017.17.3.391
  8. Li, S and C. Li. 2010. Water body extraction from Landsat ETM+ imagery using adaboost algorithm. 18th International Conference on Geoinformatics, Beijing, China, June. 18-20, pp.1-4.
  9. Ministry of Land, Infrastructure and Transport. 2013. List of rivers in Korea, pp.1-107..
  10. Ouchi, K. 2004. Principles of synthetic aperture radar for remote sensing. Tokyo Denki University Press, Tokyo, pp.133.
  11. Pulvirenti, L., M. Chinib, N. Pierdiccaa, L. Guerrieroc, and P. Ferrazzolic. 2011. Flood monitoring using multi-temporal COSMO-SkyMed data: Image segmentation and signature interpretation. Remote Sensing of Environment 115(4):990-1002. https://doi.org/10.1016/j.rse.2010.12.002
  12. Raschka, S. and V. Mirjalili. 2019. Python Machine Learning: Machine Learning and Deep Learning with Python, scikit-learn, and TensorFlow 2. Packt Publishing Ltd, Birmingham, pp.153.
  13. Sghaier, M. O., S. Foucher and R. Lepage. 2016. River extraction from highresolution SAR images combining a structural feature set and mathematical morphology. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 10(3), 1025-1038. https://doi.org/10.1109/JSTARS.2016.2609804
  14. Tonmoy, G., L. Li and J. Chakareski. 2018. Effective deep learning for semantic segmentation based bleeding zone detection in capsule endoscopy images. 25th IEEE International Conference on Image Processing(ICIP), Athens, Greece, Oct. 7-10, pp.3034-3038.
  15. Yang, I., D. Lee, T. D. Acharya and D. Han. 2018. AW3D30 and SRTM30 Digital Elevation Models - comparison around coastal islands in Gyeongsangnam-do, South Korea. Journal of the Korean Association of Geographic Information Studies 21(2):34-43. https://doi.org/10.11108/KAGIS.2018.21.2.034