基于代数多重网格的天波超视距雷达跨尺度地海杂波识别方法

Skywave over-the-horizon radar (OTHR) plays a key role in early warning of long-range targets. Coordi⁃ nate registration based on OTHR land/sea clutter recognition uses the land/sea clutter recognition results to construct the land/sea boundary or terrain contour, which is matched with the a priori geographic information to provide coordinate regis⁃ tration parameters for target positioning, which can improve the target positioning accuracy of OTHR. To meet the require⁃ ments of different types of target detection, OTHR usually adopts different signal time-widths and coherent integration points, so that the land/sea clutter spectorum data has multi-resolution and multi-scale characteristics. Designing classifiers for different resolution/scale land/sea clutter spectrum data has the problems of unbalanced training data and high mainte⁃ nance costs. Based on the idea of algebraic multigrid and interpolation-dependent image downsampling, this paper establish⁃ es the algebraic relationship between the land/sea clutter spectrum data of different scales and proposes a cross-scale deep convolution neural network land/sea clutter classifier. This method allows us to use trained low-resolution land/sea clutter classifiers to classify high-resolution data. The classification accuracy is no less than 88.26%. It also allows us to use a trained high-resolution land-sea clutter classifier to classify low-resolution data. The classification accuracy is no less than 92.53%. It is not necessary to separately design classifiers for spectrum with different scales.