12-16, 15:15–15:30 (Asia/Bangkok), Auditorium Hall 2
In recent years, the availability of geospatial data has significantly increased, with aerial photographs and Digital Elevation Models (DEMs) becoming widely accessible as open data. Additionally, the acquisition of high-resolution image data through drones has become more feasible and commonplace. However, extracting meaningful information from these vast datasets remains a labor-intensive process, often requiring significant time and resources.
While various deep learning techniques have been employed to address this challenge, they typically demand extensive effort in collecting and preparing training data. In light of these constraints, the Segment Anything Model (SAM) has emerged as a promising solution. SAM, a recent development in the field of fundamental models, offers the advantage of zero-shot classification without the need for specific training. Moreover, its Apache-2.0 license ensures accessibility for a wide range of applications.
This presentation aims to demonstrate the potential of SAM in the realm of geospatial information processing. We will showcase practical applications of SAM in analyzing geospatial data, with a focus on two critical areas: landslide detection and forest canopy mapping. These case studies will illustrate how SAM can efficiently process and extract valuable insights from complex geospatial datasets, potentially enhancing the efficiency and effectiveness of our approach to environmental monitoring and disaster risk assessment.
