Efficient Recurrent Neural Network for Classifying Target and Clutter: Feasibility Simulation of Its Real-Time Clutter Filter for a Weapon Location Radar

Abstract

The classification of radar targets and clutter has been the subject of much research. Recently, artificial intelligence technology has been favored; its accuracy has been drastically improved by the incorporation of neural networks and deep learning techniques. In this paper, we consider a recurrent neural network that classifies targets and clutter sequentially measured by a weapon location radar. A raw dataset measured by a Kalman filter and an extended Kalman filter was used to train the network. The dataset elements are time, position, radial velocity, and radar cross section. To reduce the dimension of the input features, a data conversion scheme is proposed. A total of four input features were used to train the classifier and its accuracy was analyzed. To improve the accuracy of the trained network, a combined classifier is proposed, and its properties are examined. The feasibility of using the individual and combined classifiers as a real-time clutter filter is investigated.