Presentation Information
[B-6-65]Uncertainty-Aware Semantic Communication for Drone-Based Survivor Search
◎Takumi Maeda1, Ryoichi Kawahara1 (1. Toyo Univ.)
Keywords:
semantic communication
Semantic communication has attracted considerable attention as a communication paradigm that reduces communication cost by transmitting only semantically important information. Previous studies have proposed a semantic communication framework in which semantic latent features extracted by an Unmanned Aerial Vehicle (UAV) are compressed and continuously transmitted to a server. However, in surveillance tasks such as survivor search, it is not always necessary to apply the same transmission strategy to every detection result. Therefore, this study proposes an uncertainty-aware semantic communication framework for drone-based survivor search that adaptively switches the transmission method according to the confidence of object detection. Specifically, only metadata are transmitted for highly confident detections, whereas a Region of Interest (ROI) image or semantic latent features are transmitted only for uncertain detections. This approach aims to reduce communication cost while maintaining survivor detection performance.