Author(s): Keisuke Yoshida; Shijun Pan; Takashi Kojima

Linked Author(s): Shijun Pan, Keisuke Yoshida

Keywords: DeepLabV3+; Land Cover Classification; Segment Anything Model; UAV

Abstract: Riparian land cover classification (LCC) is an important task for environmental monitoring and management, as it provides information on the state and characteristics of riparian zones, which are the transitional areas between terrestrial and aquatic ecosystems. However, LCC is challenging due to the high diversity and complexity of riparian vegetation and land use. In this paper, we evaluate the performance of Segment Anything Model (SAM), a recently proposed deep learning model for general image segmentation, for riparian LCC from aerial imagery. SAM can produce high quality object masks from input prompts such as points or boxes, and can generate masks for all objects in an image without fine-tuning. The authors compared SAM with the results derived from the DeepLabV3+ model trained by the dataset collected in Japan. Our experiments showed that SAM achieved competitive or superior results compared to other models, and can segment various riparian land cover types with high accuracy and consistency. The authors concluded that SAM is a promising tool for riparian LCC from aerial imagery, and suggest some directions for future research combining with the other model.

DOI: https://doi.org/10.3850/iahr-hic2483430201-347

Year: 2024