FMICW Radar Target Classification By Neural Network

Abstract

This document describes automatic classification of targets detected by the FMICW radar. These targets are counted and sorted to three groups (incoming, outgoing and static targets). We derived this information from the output of the neural network which marked the targets in 2D spectrum. The additional neural network has five layers. The first layer is used for the suppression of the targets with even numbers of points, which causes problems during the symmetry detection. The second and third layers detect the symmetry in the dimension (vertical or horizontal). The fourth layer checks out if the symmetry is in both dimensions and if the detection is not a false alert caused by the constellation of the targets. The fifth layer contains only 4 neurons and this layer is used for counting of the targets and classification of the targets (if they are static, incoming or outgoing). The neural network is composed of a simple block for the easy implementation on the FPGA.