Study On Feature Extraction And Evaluation Method Of Radar Emitter Signals

In modern warfare, radar is important equipment to achieve military superiority. Radar reconnaissance is one of the main content in radar countermeasure. It has already become the "eyes" and "ears" of the commander at all levels, which could help them to discriminate the radar type, function and deployment of the enemy, and could help them to know the branch of the armed forces, or even to know the troops and weapons deployment of the enemy.In signal processing of radar reconnaissance system, the identification of radar emitter signal is one of the crucial aspects. The recognition level has become an important symbol to measure the technology level of the radar countermeasures equipment. With the wide application of radar and with the degree increasing of radar electronic countermeasures, the long-term used method based on the traditional five parameters is not suitable for RES recognition in today’s radar signal environments, which have the attributes of high density, high complex waveforms and broadband spectrum agility. It is difficult to achieve satisfactory recognition effect. Therefore, the domestic and foreign scholars have done a lot of in-depth and systematic study in this field of the RES recognition. In these studies, there is a large part on the research of RES feature. They are trying to find new characteristic parameters to make up for the shortcomings of traditional five parameters. At present, there are dozens of new proposed features. However, the comprehensive effectiveness of these features is not clear. How to choose these new features is a challenge for us.Therefore, aiming at the key theoretical issues to be solved in the evaluation of RES features, this paper makes a systematic and in-depth research in the RES feature extraction and analysis, feature evaluation index system, evaluation model and algorithm design etc.. The main contributions of this paper are the following four aspects1. The extraction methods of RES feature (viz. wavelet gray moment feature, EMD energy entropy feature and texture feature) are proposed. First, the wavelet coefficients were extracted by using the continuous wavelet transform in the RES wavelet gray moment feature extraction. In addition, through experimental analysis, the choice problem of wavelet decomposition scale and continuous wavelet were solved. Then, image recognition technology was integrated into the RES feature extraction. Finally, extracting of wavelet gray moment features and texture features of RES were realized and the effect of noise on the feature and recognition ability was analyzed in detail. In the extraction of EMD energy entropy method, RES are first noise reduction through the wavelet packet decomposition and reconstruction of RES. Then through the empirical mode decomposition algorithm, RES is decomposed, the different intrinsic mode functions were obtained. The energy of the main intrinsic mode functions were extracted to form the feature vector by defining the energy of intrinsic mode function components. Finally, the typical RES were identified through the construction of classifier, and the impact of noise on the feature extraction and feature recognition capabilities were analyzed in detail.2. The separability evaluation method of RES feature was proposed based on four kinds of measure. Robustness measure index was designed and the evaluation index system of RES features was perfect. The evaluation index system of RES feature was studied deeply from the separability, complexity and robustness of three different angles. The RES recognition was studied based on the separability evaluation. The spatial distribution problem of RES feature vectors is studied. Simulation experiment results show that the multi-dimensional feature vector of RES feature is not the standard normal distribution, even if one-dimensional characteristic parameter of RES feature is consistent with the normal distribution. This conclusion provides theoretical support for the following research on comprehensive evaluation model and algorithm.3. The feature evaluation model of RES was proposed based on set pair analysis and fuzzy analytic hierarchy process. Based on the establishing of evaluation index system of RES feature, RES features were evaluated by using analytical hierarchy process. So, the analytic hierarchy process (AHP) evaluation model for RES features is proposed. Nevertheless, the AHP evaluation model cannot deal with expert fuzzy evaluation problem. So, the fuzzy AHP evaluation model of RES feature was proposed through the introduction of fuzzy theory. Finally, for the subjective problem existing in the evaluation decision of FAHP evaluation model, the evaluation model based on set pair analysis and fuzzy analytic hierarchy process was proposed by the reference of the theory thought of set pair analysis and the establishment of decision matrix. Moreover, the objective evaluation of RES features was realized.4. The feature evaluation model of RES, based on swarm intelligence, is proposed. The AHP evaluation method has subjectivity problem in the establishment of judgment matrix, so, the comprehensive evaluation model of RES feature was established based on differential evolution algorithm, particle swarm optimization algorithm and projection pursuit algorithm. Firstly, the improved differential evolution particle swarm optimization (DEPSO) algorithm was obtained through the combination of two algorithms. Then, this algorithm was used to optimize the projection direction of projection pursuit algorithm. Finally, the evaluation model of RES feature based on differential evolution algorithm, particle swarm optimization and projection pursuit algorithm is established. From the method, this model overcomes the subjectivity problem that lies in traditional AHP-based evaluation model. Moreover, a new idea for the objective comprehensive evaluation of RES feature was opened.