Segmentation of Spectrum Map for HFSWR Based on Feature Extraction

When targets fall in different types of clutter and noise background in High Frequency Surface Wave Radar (HFSWR), uniform detection algorithms always result in poor detection and tracking performances. To solve the problem, two methods are proposed in this paper by extracting the background features according to the probability distribution and the image characteristic of the spectrum map. The first method uses Parzen window for estimating probability density function and Kullback-Leibler convergence for evaluating the difference between the Azimuth-Range-Doppler cells in spectrum map which can be used to distinguish the homogenous detection background from the non-homogenous. The second method is proposed based on the image edge feature in multi-scale space. The real data results about HFSWR show that the proposed methods can not only identify noise/clutter but also help choose appropriate detection/tracking algorithms to improve performances in such complex environments.