A Novel Algorithm for Automatic Ship and Oil Spill Detection Based on Time-Frequency Methods
Marivi Tello(1) , Carlos Lopez-Martinez(1) , and Jordi Mallorqui(1)
Universitat Politecnica de Catalunya,
D3-Campus Nord - C/ Jordi Girona 1-3 ,
A novel method for automatic ship detection has been recently presented based on the wavelet transform (WT). The results obtained point out the potential of the use of a multiresolution time-frequency framework for the analysis of SAR imagery.
In fact, the scenarios in which the application of SAR techniques is especially well suited involve natural elements that can be thought of as stochastic non stationary processes. It has already been shown that, as it can perceive a structure in the context of its surroundings, the human vision can manage those non stationarities in a convenient way, overcoming existing automatic algorithms in extracting features in a complex scene: for example, the human eye is superior in observing a slick in the context of the surrounding sea and, surprisingly, some vessels undetected by conventional techniques are visible by eye. Also, SAR imagery is affected by different kinds of disturbances, such as speckle and a great number of discontinuity effects. Nevertheless, a human operator can be able to intelligently manage those heterogeneity sources thanks to his capability of perceiving images in a multiscale way. And, as a matter of fact, as multiresolution processing is able to model the operation of the human vision, it seems interesting to axe the interpretation of SAR images by means of time-frequency methods and in particular, by means of the wavelet tools.
In addition to previously published results on ship detection, this paper will show that these considerations still remain valid for oil spill detection. A novel algorithm whose objective is to provide reliable both ship detection and extraction of oil spills and slicks in SAR imagery will then be presented, justified and tested on ENVISAT imagery (available ground-truth for vessels).
More specifically, a recent research on the analysis of oil slicks in SAR imagery has revealed an interesting observation: even if oil slicks and wind speed decrease areas (principal source of look alikes) present a similar appearance in oceanic SAR images, they can be discriminated through higher order moments. Based on this assumption, the main objective of this work is to provide a precise measure of local regularity of the SAR image in order to perform a robust segmentation. It can be shown that the local regularity of signals is characterized by the decay of the WT amplitude across scales, since this decay is related to the uniform and point wise Lipschitz regularity of the signal. This property has been used in textural analysis in a qualitative point of view: singularities and edges are detected by following the WT local maxima across scales. In the scope of SAR, it is proposed to perform a multiscale segmentation in SAR imagery, building on the idea of characterizing and exploiting the scale-to-scale statistical variations due to radar speckle.