Automatic Extraction and Analysis of Shipping Patterns from Visible Satellite Imagery

This paper gives an overview of a program to automatically extract ship positions and sizes from a satellite image. This is based on the extraction and classification of objects in the image which may correspond to ships and the sea surface. Shiplike objects which are enclosed by sealike objects are extracted and catalogued as ships-underway. To extract the ship and sea objects rule based processing of the binary object forest representation of the image is used, The example shown is a noisy low contrast SPOT panchromatic image showing several small ( 10-20m) boats, some of which are in a manna and some in both deep and shallow water. Its is shown that detection rates of greater than 70% (without false alarms) can be obtained for the smaller boats and considerably higher for the larger boats.