Deep Learning for Iceberg Detection in Satellite Images
Iceberg roots a dangerous risk for transportation, marine oil and gas production plants. Data on Iceberg and Island is significant for atmosphere science and for different marine activities in the sea. The detection and monitoring of the icy objects in the polar zone, which is often dark and cloudy are done with the well-established tool know as synthetic aperture radar (SAR). In this paper, the CNN method will be utilized for distinguishing the iceberg from high-resolution satellite images. This paper presents a robust and efficient method of extraction of features for SAR target classification using deep features. This approach is centered on the concept of adaptability and focuses on making better efficiency of iceberg detection in unclear environmental situations with broad variation in textural, scale, and shape. The procedure used for iceberg classification includes SAR images, deep features extracted based on Threshold and CNN classifier for classification of the iceberg and other objects. The CNN procedure is used on a dataset that contains high-resolution SAR images that were captured from satellites Landsat 8 and Sentinel 1 (Southern Ocean).
Keywords- Iceberg detection, CNNs, SAR images