Earticle

초록

영어

This study aimed to determine optimal conditions for Watershed algorithm-based individual tree crown segmentation using UAV-derived Canopy Height Model (CHM) and grayscale imagery. Three sample plots were established in coniferous forests (Pinus koraiensis and Larix kaempferi) and a broadleaf forest (Quercus mongolica) in Pocheon-si, Gyeonggi-do, and Cheongju-si, Chungcheongbuk-do, South Korea. CHM and grayscale images were constructed at spatial resolutions of 5 cm, 25 cm, and 50 cm. Gap regions were classified using three thresholding methods (Otsu, Entropy, and Percentile 5%, 10%, 15%), followed by individual tree crown segmentation using the Watershed algorithm. Accuracy was assessed by comparing segmentation results with reference data generated through visual interpretation of the imagery. For gap region classification, the Percentile 10% method achieved the most balanced accuracy between crown and gap regions, with a mean crown IoU of 0.88, and showed consistent performance across tree species, data types, and spatial resolutions. For Watershed-based individual tree segmentation, the MAE of individual tree counts showed the most consistent segmentation accuracy at a spatial resolution of 25 cm in Pinus koraiensis and Larix kaempferi, with 50 trees/ha for CHM and 63 trees/ha for grayscale images. For Quercus mongolica, segmentation accuracy was highest at a spatial resolution of 50 cm, with 50 trees/ha for CHM and 100 trees/ha for grayscale images. Over-segmentation occurred at 5 cm, while under-segmentation was observed at 50 cm. CHM exhibited higher accuracy at finer resolutions, whereas grayscale imagery performed better at coarser resolutions. Regarding tree species, coniferous forests with conical crown structures showed higher segmentation accuracy than broadleaf forests. This study highlights the need to select appropriate data types and spatial resolutions according to forest stand characteristics and is expected to contribute to UAV-based individual tree-level forest structure analysis and precision forest inventory.

목차

Abstract
Introduction
Materials and Methods
Study area and data
Research method
Data acquisition and preprocessing
Gap region classification using different methods and thresholds
Individual tree crown segmentation using the watershed algorithm
Result and Discussion
Gap region classification accuracy by thresholding method
Individual tree crown segmentation results by spatial resolution and data type
Conclusion
Acknowledgements
References

키워드

저자

• Ah-Ri Shin [ Department of Forest Management, College of Forest and Environmental Sciences, Kangwon National University, Chuncheon 24341, Republic of Korea ]
• Woo-Dam Sim [ Department of Forest Management, College of Forest and Environmental Sciences, Kangwon National University, Chuncheon 24341, Republic of Korea ]
• Sang-Jin Lee [ Department of Forest Management, College of Forest and Environmental Sciences, Kangwon National University, Chuncheon 24341, Republic of Korea ]
• Jung-Soo Lee [ Department of Forest Management, College of Forest and Environmental Sciences, Kangwon National University, Chuncheon 24341, Republic of Korea ] Corresponding Author

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  Journal of Forest and Environmental Science [산림과학연구]

• 간기

  계간

• pISSN

  2288-9744

• eISSN

  2288-9752

• 수록기간

  1981~2026

• 등재여부

  KCI 등재

• 십진분류

  KDC 526 DDC 634