An improved algorithm for ice shelf crevasses extraction considering morphological characteristics

The task of crevasses segmentation in remote sensing images plays an important role in the fields of crevasses change monitoring and ice shelf stability study. In order to improve the fragmentation problem of the current crevasses segmentation results, this paper improves the high-precision extraction method of crevasses by taking into account the morphological characteristics of the zigzag and staggered distribution of crevasses. The experimental results show that the improved algorithm improves the F1-score, MIoU and Recall by 1.01%, 1.69% and 2.43%, respectively, compared with the traditional algorithm, which effectively improves the connectivity of the crevasses extraction results, and it has the potential to be applied in crevasses segmentation scenarios based on remotely sensed images.