[3Xin4] Poster session 1

[3Xin4-01] Application of Interactive Semantic Segmentation Using a Human-Centered AI System to Remote Sensing Imagery

〇Takayuki Shinohara¹, YongHe Li¹, Mitsuteru Sakamoto¹, Toshiaki Satoh¹ (1.Pasco Corporation)

Keywords:Human Centric AI, segmentation, remote sensing

Remote sensing images are a valuable resource for extracting geographical features, but the process of visually deciphering these images can be time-consuming and error-prone. In recent years, researchers have attempted to improve the efficiency of this process through the use of fully automatic methods based on deep learning. While these methods have shown promise, they have not yet reached the level of accuracy and effectiveness of human visual interpretation.
To address this issue, we propose an interactive approach to extracting geographical features from remote-sensing images using AI. Rather than relying on fully automatic extraction, our method involves a human operator clicking on a specific pixel or region of interest. The AI system then performs object detection and instance segmentation in the vicinity of the clicked pixel, allowing for a more targeted and accurate extraction process. We achieved high performance in tasks where fully automatic extraction methods may struggle.