Radar Target Classification Using Bi-Lstm Model Of Recurrent Neural Networks

Abstract

The classification of targets is one of the challenging tasks in the field of modern radar systems. Manual feature extraction with high level computer vision algorithms or neural networks needs knowledge of the subject domain.  In this explore a pretrained Recurrent neural network (RNN) with a bidirectional LSTM model for radar target classification where a separate feature extraction is not required for this network. The canonical models for simple geometric shapes, namely sphere, cylinder, disc and frustum and complex geometric shapes, namely complex sphere, complex cylinder, complex frustum and complex disc  are developed using dedicated phased backscatterd algorithms. In terms of performance metrics, Bi-LSTM network gives an accuracy of around 99.77% for simple targets and 99.63% for complex targets, far better than the machine learning models. The experimental study on target classification of radar sequence data using Bidirectional Long short-term memory (Bi-LSTM) of RNN is presented.