Recognition multifrequency microwave images of simple objects behind dielectric wall using neural networks

Abstract

Through-the-wall radars development is very actual problem. For instance, the equipment of such sort can be used for testing the state of valve situated behind a wall. Using wideband microwave signals of millimeter range allow us to detect objects positioned behind barrier non-transparent in optical range and to obtain their radio images. This equipment operates in the frequency band of 38 - 52 GHz. The pyramidal horn as radiating and receiving antenna is used. The main idea of corresponding approach is the combination of the synthesis of time-domain (spatial) signal from frequency data for obtaining longitudinal information along y-axis and scanning for obtaining transversal one along x-axis. Thus, 3-D radio image with data about insert reflection along z-axis can be formed. The advantage of this approach in contrast to single-frequency signals is possibility of separation of reflection from the object under consideration and the wall. As result, the image of the object similar to the one for the object in free space can be obtained. The purpose of the paper is to investigate possibility to recognize groups of objects situated behind dielectric barrier at the various distances. The objects under recognition have been metal parallelepipeds and cylinders with variation of sizes. The bases of the parallelepipeds were 25, 45 and 55 mm. The diameters of the cylinders were 25, 45 and 75 mm. The height of objects was 110 mm. Object recognition has been implemented using neural networks and correlation techniques. As the result of the study, it has been found the best settings for the recognition of objects of different geometric shapes.