Using LeNet-5 and AlexNet Architectures in Deep Learning Approach to Detect and Classify LPI Radar Signals

Low probability of intercept (LPI) radars are difficult to detect and identify by electronic intelligence receivers due to their low power, wide bandwidth and frequency variability. With the emergence of this technology, new methods of signal and image processing are constantly required to first identify, then classify, and finally extract the characteristics of these radar signals. To solve the problem, today deep learning is an important technical method in the signal and image processing fields. Through using this method, this paper will investigate the possibility of detecting and classifying different signals of LPI radars. To do this, using Short-Time Fourier Transform (STFT), we will analyze the received signal in the time-frequency domain, and then to detect and classify the LPI radar signal waveforms we send the output, in image format, to the AlexNet and the LeNet deep convolutional neural network (CNN) models. The simulation results show that, in SNR=-5dB, the accuracy of the AlexNet and the LeNet methods are 97.34% and 94% respectively, indicating the better performance of the AlexNet method.