CLASSIFICATION OF POLAR SATELLITE DATA USING IMAGE FEATURES AND DECISION TREE CLASSIFIER

• Main Authors: ムラモト ケンイチロウ, ヤマノウチ タカシ, Kenichiro MURAMOTO, Takashi YAMANOUCHI
• Format: Report
• Language: English
• Published: Department of Electrical and Computer Engineering, Faculty of Engineering, Kanazawa University 1996
• Subjects: Antarc*; Antarctica; Polar meteorology and glaciology; Proceedings of the NIPR Symposium on Polar Meteorology and Glaciology; Sea ice
• Online Access: https://nipr.repo.nii.ac.jp/?action=repository_uri&item_id=3932; http://id.nii.ac.jp/1291/00003932/; https://nipr.repo.nii.ac.jp/?action=repository_action_common_download&item_id=3932&item_no=1&attribute_id=18&file_no=1

In the polar region, it is difficult to discriminate between clouds and ground surface from satellite visible or infrared data, because of the high albedo and low surface temperature of snow and ice cover. In this paper, a method to classify clouds, sea ice and ground is proposed. This study is based upon analysis of the NOAA/AVHRR infrared images in Antarctica. The algorithm consists of two major approaches : estimating image features and a classification algorithm. A decision tree classifier is designed to classify the region into one of three classes using six image features. Though sea ice and ground can be largely separated using only one feature, more than three features are necessary to separate clouds.