As a mid-level vision task, water segmentation is essential for unmanned surface systems such as to get the sailing area and avoid the floating objects to ensure vehicles safety in maritime environment. However, existing deep learning method of water segmentation has a little weakness on maintaining origin-scale fineness. At the same time, our intuition is that ground truth is a powerful tool for optimizing the global network. In this paper, we give solutions through these two aspects using superior guidance and aligned fusion strategy. Aligned fusion provides symmetric and fine-grained fusion strategy when using encoder-decoder structured network, and superior guidance enforce the segmentation network grasp the discrimination criterion much earlier before output part of the network. Both of them show benefits for water segmentation of unmanned surface system's image data. The experimental results on Marine Semantic Segmentation Training dataset show its effectiveness and efficiency.
Water Segmentation with Superior Guidance and Aligned Fusion Strategy for Unmanned Surface Vehicles in Maritime Environment
2022-10-28
682030 byte
Conference paper
Electronic Resource
English
The Operation of Unmanned Air Vehicles in the Maritime Environment
British Library Conference Proceedings | 1993
|The Operation of Unmanned Air Vehicles in the Maritime Environment
British Library Conference Proceedings | 1993
|Guidance, Navigation and Control of Unmanned Surface Vehicles
Tema Archive | 2015
|Guidance of Unmanned Aerial Vehicles
TIBKAT | 2011
|