Computer-based identification and tracking of Antarctic icebergs in SAR images

Icebergs play an important role in climate through the transfer of fresh water and heat between ice sheets and the oceans. Synthetic aperture radar (SAR) images offer a robust way of observing icebergs in the often dark and cloud-covered polar areas. A computer-based technique is proposed to identify icebergs at higher resolution than in the past, allowing the shape of small to medium icebergs to be retained and used for tracking between images acquired at different times. The identification consists of (i) automatic segmentation, followed by (ii) classification of these objects as icebergs or nonicebergs. The classification stage requires user intervention as the icebergs are not always a separable class from other objects, most notably sea ice. The icebergs are then matched between images using their size and shape similarities, and their position tracked. The system performance was assessed by applying the technique to three wintertime ERS-1 PRI images, around Kapp Norvegia (10°W), on the Eastern Weddell Sea coast. The results show both the identification and the tracking to be effective.