Space Target Anomaly Detection Based on Gaussian Mixture Model and Micro-Doppler Features

With the dramatic increase in human space activities, anomaly detection becomes an important issue in passive space target surveillance. In this article, an anomaly detection algorithm based on the Gaussian mixture model (GMM) and radar micro-Doppler features is proposed to detect the abnormal motion status of the space target. By coherent sampling and time-frequency (TF) analysis on the radar echo with additive white Gaussian noise (AWGN) corresponding to the normal motion statuses of the target, four micro-Doppler features are extracted and tested for normal distribution. Furthermore, the distribution of the multidimensional features and the corresponding parameters are fit and estimated by the GMM and expectation-maximization (EM) algorithm. Then, an anomaly detector is derived by solving for the decision region using the fit probability density function (pdf) and a preset confidence level. Experimental results show that the average anomaly detection rate of the proposed method is 16.7%, 19.1%, and 34.0% higher than the one-class support vector machine (OCSVM), the convex hull, and the convolutional autoencoder (CAE)-based methods, respectively.