Sea Ice Change Detection from Synthetic Aperture Radar Images Based on Self-Paced Boosting Learning

The change information of sea ice is important for navigation safety and natural resource extraction, and polar sea ice monitoring has drawn increasing attentions. In this paper, we focus on the problem of sea ice change detection from Synthetic Aperture Radar (SAR) images. Existing works usually treat the problem as a classification task. Recently, researchers noticed that humans and animals learn much better when the examples are organized in a meaningful order, and proposed Self-Paced Boost Learning (SPBL) to simulate the process. Inspired by this, we proposed a novel sea ice change detection method based on SPBL. First, log-ratio operator is utilized for difference image generation. Then, a hierarchical fuzzy c-means (FCM) clustering algorithm is employed to select reliable samples. Finally, based upon these samples, SPBL is employed to classify pixels from the original SAR images into unchanged and changed class. The quantitative and qualitative analysis on two real SAR datasets has demonstrated the effectiveness of the proposed method.