Automated Sea Ice Segmentation (ASIS)

The authors describe a sea ice analysis tool called ASIS (Automated Sea Ice Segmentation). This tool integrates image processing, data mining, and machine learning methodologies to determine the number of visually separable classes in ERS and RADARSAT sea ice images. First, it performs dynamic local thresholding to obtain local details and preserve global information throughout the image. Second, it utilizes a data discretization or image quantization scheme to obtain significant classes of the image through multiresolution blurring and tracking. Third, it computes spatial attributes of each class and subjects the information to an unsupervised clustering technique based on the Aggregated Population Equalization (APE) concept. This concept self-organizes the population of classes by promoting the aggregation of different classes to obtain an equilibrium of population strengths within the environment and encouraging disintegration of any over-diverse population. The authors have designed ASIS as a pre-processor to help analyze sea ice images as well as to provide a basis for human classification of sea ice types that it identifies. Therefore, they have also implemented a JAVA-based graphical user interface that facilitates the computer-human interaction for this whole system. They have tested ASIS on more than 300 ERS-1, ERS-2 and RADARSAT SAR sea ice imagery. In general, the results are satisfactory. They have also conducted both qualitative and quantitative evaluations of ASIS and found that the segmentation classes correspond well to visually identifiable sea ice classes.