Sea-Sky Line Detection via an Ensemble Semantic Segmentation and Hough Transform Method

Sea-sky line detection plays an important role in the autonomous navigation of unmanned surface vessel. However, the detection of sea-sky line is easily affected by sea waves, ships, weather, and other factors, resulting in inaccurate detection results. In this paper, the improved Mo-SegNet image semantic segmentation algorithm based on the SegNet network is selected to segment the ocean image to effectively suppress noise interference. Finally, Hough transform is used to fit the edge pixels of sea and sky, and the detection results of sea-sky line are obtained. This paper compares the detection results of this model under different semantic segmentation algorithms. The experimental results show that the Mo-SegNet network has high accuracy and robustness, and the accuracy of the network reaches 99.31% on self-built and labeled dataset.