Han Jianfeng, Gao Xiongwei, Song Lili, Fang Jiandong, Tao Yongzhao, Deng Haixin, Yao Jie
School of Information Engineering, Inner Mongolia University of Technology, Hohhot 010080, China.
Inner Mongolia Key Laboratory of Intelligent Perception and System Engineering, Hohhot 010080, China.
Sensors (Basel). 2025 Apr 20;25(8):2600. doi: 10.3390/s25082600.
Water surface inspection is a crucial instrument for safeguarding the aquatic environment. UAVs enhance the efficiency of water area inspections due to their high mobility and extensive coverage. This paper introduces two UAV inspection methodologies for the characteristics of rivers and lakes, along with an efficient semantic segmentation algorithm, WaterSegLite (Water Segmentation Lightweight algorithm), for UAV visual navigation. The algorithm employs the UAV's aerial perspective alongside a streamlined neural network architecture to facilitate rapid real-time segmentation of water bodies and to furnish positional data to the UAV for visual navigation. The experimental findings indicate that WaterSegLite achieves a segmentation accuracy (mIoU) of 93.81% and an F1 score of 95.44%, surpassing the baseline model by 2.7% and 2.23%, respectively. Simultaneously, the processing frame rate of this algorithm on the airborne device attains 28.27 frames per second, fully satisfying the requirements for real-time water surface inspection by UAVs. This paper offers technical assistance for UAV inspection techniques in aquatic environments and presents innovative concepts for the intelligent advancement of water surface inspection.
水面检测是保护水生环境的一项关键手段。无人机因其高机动性和广泛的覆盖范围，提高了水域检测的效率。本文针对河流和湖泊的特点介绍了两种无人机检测方法，以及一种用于无人机视觉导航的高效语义分割算法——WaterSegLite（轻量级水域分割算法）。该算法利用无人机的俯瞰视角以及简化的神经网络架构，以实现水体的快速实时分割，并为无人机视觉导航提供位置数据。实验结果表明，WaterSegLite的分割精度（平均交并比）达到93.81%，F1分数达到95.44%，分别比基线模型高出2.7%和2.23%。同时，该算法在空中设备上的处理帧率达到每秒28.27帧，完全满足无人机实时水面检测的要求。本文为水生环境中的无人机检测技术提供了技术支持，并为水面检测的智能发展提出了创新理念。
