Multiscale algorithm for ship detection in mid, high and very high resolution optical imagery

In the context of improving the responsiveness of a maritime surveillance service, we propose a ship detection algorithm in optical satellite imagery. The major objective is the detection of vessels of any sizes in any resolution images. Signatures of ships and false alarm sources are analyzed and the state of the art studied. The wavelet transform is used for extracting at different scales details corresponding to vessels. At each scale, details coefficients are processed to provide an improved signal to noise ratio data that can be segmented by a conventional adaptive thresholding method. Contours from these segmentations are assembled in trees which are then processed in the discrimination step. From one scale to the other changes in the contour of a same target are analyzed to determine if it corresponds to a vessel. The detection capabilities at the base scale are improved, and application to other scales allows the detection of any size of vessel.