KDP-Net: An Efficient Semantic Segmentation Network for Emergency Landing of Unmanned Aerial Vehicles

• DOI: 10.3390/drones8020046
• Publication Date: 2024-02-01T16:25:55Z
• Authors: Zhiqi Zhang; Yifan Zhang; Shao Xiang; Lu Wei
• Keywords: Aerial image

As the application of UAVs becomes more and widespread, accidents such as accidental injuries to personnel, property damage, loss destruction due UAV crashes also occur in daily use scenarios. To reduce occurrence accidents, need have ability autonomously choose a safe area land an situation, key lies realizing on-board real-time semantic segmentation processing. In this paper, we propose efficient method called KDP-Net for characteristics large feature scale changes high processing requirements during emergency landing process. The proposed KDP module can effectively improve accuracy performance backbone network; Bilateral Segmentation Network improves extraction speed important categories training phase; edge classification fine features. experimental results on UDD6 SDD show that reaches 85.25 fps 108.11 while mIoU 76.9% 67.14%, respectively. 53.72 38.79 when measured Jetson Orin, which meet airborne landing.