Kiwon Lee, Hansung University, Korea
Kiwon Lee received his Ph.D. degree from Seoul National University in 1995 and worked as a Senior Researcher at Korea's Electronics and Telecommunications Research Institute (ETRI) from 1995 to 2001. Since 2001, he has taught and conducted research on several subjects, including satellite image processing, geographic information systems, 3D graphics, and cloud computing, at Hansung University, Seoul in Korea. He also served the Korean Society of Remote Sensing as an executive director from 2005 to 2007, editor-in-chief from 2007 to 2009, and vice president from 2010 to 2012.
As part of his academic achievements, he received the Best Researcher Award from Hansung University in 2005, 2007, 2009, 2011, and 2016. In 2010 and 2011, he received the Best Presentation Paper Award at the Annual Conference of the Korea Spatial Information Society. Notably, in 2010, he was awarded two government honors: The Minister of Land, Transport, and Maritime Affairs Award of Korea and the Annual Best Paper Award from the Korean Federation of Science and Technology Societies. In 2011, he received the Best Paper Award from the Korean Journal of Remote Sensing, along with the Best Cited Paper Award from the same journal. He also received the Best Presentation Paper Award from the Korean Society for Geospatial Information Science, and the Most Prolific Author Award from the Korean Journal of Remote Sensing in 2017, 2023, and 2024, respectively.
(Onsite Talk) Speech Title: Application of Machine Learning Scheme on Google Earth Engine (GEE) and Segment Anything model (SAM) to Geo-morphological Feature Extraction on High-resolution Satellite Images
High-resolution satellite imagery collected daily is a crucial enormous data resource for major countries and institutions operating satellites. These images are directly utilized in various fields, such as defense and security, urban planning, precision agriculture, environmental monitoring, disaster management and recovery, and digital mapping. A typical technical requirement across these fields is the automatic extraction of objects of interest. Various object classification and segmentation techniques are continuously being developed to achieve this. This study aims to recognize and extract natural objects instead of artificial structures using Meta's Segment Anything Model (SAM), a vision transformer (ViT) model introduced in 2023. While numerous studies have demonstrated high accuracy in detecting artificial objects, such as vehicles and buildings with well-defined shapes and contours, research on natural objects like rivers and wetlands remains relatively limited. In this study, we specifically fine-tuned SAM for natural object segmentation and quantitatively compared the accuracy of results before and after fine-tuning. Additionally, we conducted another experiment comparing objects extracted by SAM with classified satellite imagery obtained through machine learning in Google Earth Engine (GEE). The study presents case studies demonstrating how different data processing techniques and applied data types influence the results. The results obtained through machine learning in GEE and the segmentation results from SAM were quantitatively compared based on the ground truth reference values for the same target objects. The analysis reveals that both SAM-based segmentation and GEE-based classification produce meaningful outcomes. Furthermore, integrating and combining data generally yields superior results compared to using a single dataset.
Zhe Sun, Guangzhou University, China
Zhe Sun is an associate professor at the Cyberspace Institute of Advanced Technology, Guangzhou University in China. He earned his B.S. degree from Dalian University of Technology in 2010, followed by an M.S. degree from the University of Science and Technology of China in 2012, and finally a Ph.D. degree from the University of Chinese Academy of Sciences in 2019. In recognition of his exceptional achievements, Zhe Sun was awarded the prestigious President's Award by the Chinese Academy of Sciences in 2019. Moreover, he serves as a committee member of the Big Data Security and Privacy Computing Committee within the Chinese Association for Chinese Information Society. He has authored over 40 publications in international conferences and journals, focusing on privacy-preserving mechanisms for multiparty data sharing and deep learning.