This paper proposes a technique for recognizing anthropogenic location objects on an underlying surface that is based on measuring the complete polarization scattering matrix (PSM) of a radiolocation scene. The intelligent analysis of a PSM implies the generation of adaptive robust estimates for trends and covariance matrices, as well as reflected non-stationary non-Gaussian signals. Based on the results of the digital computer simulation experiments and field measurements of the radiolocation scene's PSM, machine learning algorithms for clustering and classifying the elements of an underlying surface and location object are verified. The possibility of using neural network architecture in the form of a support vector machine for real-time implementation of these algorithms is substantiated.
展开▼
