| Abstract: | An automated ship detection technique includes accessing data associated
with an image of a portion of Earth. The data includes reflectance
values. A first portion of pixels within the image are masked with a
cloud and land mask based on spectral flatness of the reflectance values
associated with the pixels. A given pixel selected from the first portion
of pixels is unmasked when a threshold number of localized pixels
surrounding the given pixel are not masked by the cloud and land mask. A
spatial variability image is generated based on spatial derivatives of
the reflectance values of the pixels which remain unmasked by the cloud
and land mask. The spatial variability image is thresholded to identify
one or more regions within the image as possible ship detection regions. |