Radar Image Recognition with Simultaneous Selection of Training Subspaces

Radar Image Recognition with Simultaneous Selection of Training Subspaces

연구 분야: Artificial Intelligence

학회: Pattern Recognition and Image Analysis

A task of recognizing radar images is considered. Typically, these images have a lower resolution than, say, optical images. Therefore, the formation of features in these images, in the form of distinguishable characteristics of details of objects, presents significant difficulties. As components of the feature vectors in such cases, image samples are usually taken directly, and the feature vectors are formed by scanning image objects in rows or columns. The idea of the proposed recognition technology is, for each pair (test and reference) of images, to search for subspaces (fragments) that would be closest in the sense of a given criterion. As a proximity measure, used both to select these fragments and to decide on their belonging to a certain class, a criterion of conjugacy of the current vector with the subspaces formed by these fragments is used. Results of an experiment on radar image recognition from the open database MSTAR confirm the efficiency of the proposed approach.