Doppler spectrum segmentation of radar sea clutter by mean-shift and information geometry metric

International audienceRadar sea clutter inhomogeneity in range is characterized by Doppler mean and spectrum width variations. We propose a new approach for robust statistical density estimation and segmentation of sea clutter Doppler spectrum. In each range cell, Doppler is characterized by a Toeplitz Hermitian Positive Definite covariance matrix that is coded in Poincaré's unit poly-disk and we use adaptation of standard kernel methods to density estimation on this specific Riemannian manifold. Based on this non-parametric approach to estimate statistical density of Doppler Spectrum, we address the problem of sea clutter data mapping and segmentation by extending " Mean-Shift " tool for these densities on Poincaré's unit poly-disk. This statistical segmentation is requested for robust detection of targets in sea clutter, especially in case of high sea state. I. INHOMOGENEOUS DOPPLER SPECTRUM OF SEA CLUTTER We are studying highly variable Doppler spectrum characteristics that occur in the littoral sea clutter environment, to improve robustness of adaptive coherent CFAR detector. Doppler that is observed varies from 0 to 5 m/s and peaks at more than 10 m/s induced by breaking waves. The coastal circulation has local characteristics and is very variable depending on weather conditions. Close to the shore, we have to take into account drift currents. The permanent and seasonal components constitute the so-called general currents. These are generally low. The wind generates the drive current drift of the surface layers, which is transmitted by viscosity to the deeper layers. According to the theory, for a wind that had blown in the same direction for several days at any point of a body of water and indefinite water depth (over 200 m), the surface drift current is directed at 45 ° of the wind direction. The decrease of these parameters (duration, extent, depth) has the effect of reducing the deviation. So, near the coast and variable wind, the surface drift current is substantially oriented in the direction of the wind. The coast is an obstacle for the drift current, causing an accumulation or conversely a water withdrawal, according to the relative orientation of the wind and of the coastline. Tidal currents are often distinguished by their very high speed. They can indeed be greater than 10 knots (5 m/s). Usually the directions of the tide vary very quickly. The variation of the current amplitude with the amplitude of the wave is not equal if the tide behaves like a wave, a standing wave or an hydraulic phenomenon of filling/emptying a basin. In figure 1, we give the surface currents in knots from the SHOM close to Ouessant Island. Variations can therefore be seen from 0 to 4 m/s. Figure 1. Maximum current in knots in French Brittany close to the Shore IFREMER has developed the accurate Mars model (coastal hydrodynamic model) as illustrated in the figure 2: Figure 2. Residual streams in m 2 /s (speed x water level) for different exposure to winds, illustrating the spatial variability of the current. As consequence, Doppler mean is not zero and varies along the range axis, induced by means of Doppler surface current (which may vary depending on the distance to the coast and wind exposure). These variations may be important and equal to several meters per second (tidal current and the case of variable wind exposure in the presence of island or local phenomena depending on the water depth). The spectral width of the sea clutter may also vary along the distance axis based on the variable swell after exposure to wind. Simulation of coherent radar sea clutter has been described in [20,21,22,23]. For the Ground Clutter of the Shore, the Doppler spectrum width could vary with vegetation and wind exposure. We illustrate in the figure 3, Doppler variations of clutter in littoral environment