Deep Learning-Based Approach for Low Probability of Intercept Radar Signal Detection and Classification

Abstract

Detection and classification of Low Probability of Interception (LPI) radar signals is one of the most important challenges in electronic warfare (EW), since there are limited methods for identifying these type of signals. In this paper, a radar waveform automatic identification system for detecting and classifying LPI radar is studied, and accordingly we propose a method based on deep learning networks to detect and classify LPI radar waveforms. To this end, the GoogLeNet architecture as one of the well-known convolutional neural networks (CNN) is utilized. We employ the Short Time Fourier Transform (STFT) for time-frequency analysis in order to construct the entry image for proposed method 1,2 (improved the GoogLeNet and AlexNet networks) to recognize offline training and online recognition. After the training procedure with the supervised data sets the proposed method 1,2 can detect and classify nine modulation types of LPI radar, including LFM, poly-phase (P1, P2, P3, P4) and poly-time (T1, T2, T3, T4) waveforms. The numerical results for proposed method 1, show considerable accuracies up to 98.7% at the SNR level of –15 dB, which outperforms the existing methods.