The Identification of Ice Floes and Calculation of Sea Ice Concentration Based on a Deep Learning Method

When navigating ships in cold regions, sea ice concentration plays a crucial role determining ship’s navigability. However, automatically extracting the and floe size distribution remains challenging, due to difficulty detecting all floes from images captured complex polar environments, particularly those that include both ice. In this paper, we propose using YOLACT network address issue. Cameras installed on ship collect during transit an image dataset is constructed train model can intelligently identify targets remove any noisy targets. To overcome challenge of identifying seemingly connected floes, non-maximum suppression (NMS) improved. Binarization then applied process detection results, with aim obtaining accurate concentration. We present color map histogram associated based size. The speed calculating density each reaches 21 FPS results show be accurately measured. provide case study demonstrate effectiveness proposed approach.