2. SHIP DETECTION AND FEATURE EXTRACTION
IN SAR IMAGES
Automatic ship detection systems should be composed by,
at least, three stages: land mask estimation, target detection,
and ship identification. The objective of the land mask
estimation is to obtain a clustered image where land and sea
are clearly identified, in order to define the area of interest,
sea area. DeMSAR land mask estimation is based on a
combination of an edge detector [3] [4] and superresolution
techniques without information provided by another sensor
[5] [6]. The proposed approach doesn’t require a previous
speckle filtering process, reducing the computational cost.
Ships are typically constructed from large flat metal sheets
and, hence, they can be detected as individual pixels or
groups of pixels that are unusually bright compared to their
surroundings in SAR image. DeMSAR applies a double
parameter CFAR detector of potential ships as pre-screening
algorithm [7] and also includes a feature extractor that
allows the estimation of size, length, orientation and
position, besides the detection of associated ship wakes [8]
[9]. These extracted features along with contextual
information are used to reduce false alarms.