Radar Emitter Signal Recognition Method Based On Time-frequency Image Processing

With the rapid development of modern electronic and radar technology, the electromagnetic signal is becoming more and more intensive, and the signal density of electromagnetic threat environment can reach a million levels, at the same time new radar system has emerged and occupied a domain position gradually. The recognition of radar signals based on the conventional parameters has been unable to satisfy the increasing complex electromagnetic signal environment, so it is essential to study the characteristics of the intra-pulse parameters intensively to meet the needs of modern electronic warfare environment. Time-frequency analysis as a powerful tool for newly developed non-stationary signals can transform time-domain signal into two-dimensional time-frequency signal, which means that the frequency information of one-dimensional time-domain signal can be expressed by the time and frequency of two-dimensional signal with reflecting the time-frequency information of radar signals. In recent years, wide attention has been paid to the time-frequency distribution in the field of image processing and signal detection.In this thesis, starting from the time-frequency analysis of typical radar emitter signals, we discuss the time-frequency distribution of radar emitter signals as a grayscale images with the combination of the algorithms and tools in digital image processing technology to make a series of preprocesses of gray image. Then we extract the Hu moments and Pseudo-Zernike moments from the gray image we have got, and use the principle component analysis to reduce the dimension of the image matrix and further reduce noise. Then we adopt the algorithm of BP neural network and the extreme learning machine ELM to accomplish the purpose of the classification and identification for radar emitter signals. The main work is summarized as follows:1. First the mathematic model of radar emitter signals is given, and the two types of intra-pulse modulation are introduced, which is intentional characteristics and unintentional features. And then we make a comparison of the environment of radar emitter signals with that of radar signals, and give the concrete expressions of six typical radar emitter signals and simulate the waveforms in the time domain and frequency domain.2. The basic theories and common methods of time-frequency analysis are introduced, including STFT, WVD and Cohen's time-frequency distribution. Then we analyze the generation and suppression of the cross-term in WVD and simulate the time-frequency distribution of six kinds of radar emitter signals. Though it is not difficult to distinguish the modulation type of signals, to extract the effective identification features directly from the time-frequency distribution of signals is more difficult. Therefore, based on image processing technology, this thesis makes series of preprocesses in the images of time-frequency distribution, thereby transforming the recognition of radar emitter signals into that of the time-frequency images.3. With the method of recognizing radar emitter signals based on time-frequency analysis, we transform time-frequency images into grayscale images and make a series of preprocesses by use of the algorithms and tools in digital image processing technology. After that, we extract the Hu moments and Pseudo-Zernike moments from the gray images we have got and use the principle component analysis to reduce the dimension of the image matrix and further reduce noise.4. For the classification of radar emitter signals, this thesis based on the feature extraction of time-frequency images raises the algorithm of BP neural network and the extreme learning machine ELM to classify the same modulation types and different modulation types of radar emitter signals. As the simulation results show, it can be more accurate to classify and identify the radar emitter signals according to the time-frequency features of each radar emitter signal.