关键词: 辐射源个体识别, 特征提取, 小波基, 度量学习, 信号表示

Abstract: In response to the variability of individual emitters and the complexity of the electromagnetic environment, radar emitter identification faces challenges such as slow feature extraction and low identification accuracy. This paper proposes a method for specific emitter identification of multi-feature extraction and metric network under dynamic wavelet bases selection. Initially, wavelet bases are dynamically selected for emitter signals using the principle of sparsity, converting the original signal into various signal representations based on the energy of node coefficients. Subsequently, multi-scale statistical features are extracted to form a set of feature vectors. Finally, an improved Mahalanobis distance metric learning algorithm is employed to optimize the feature space, in conjunction with a feedforward neural network for classification and identification. Experimental results demonstrate that the proposed method achieves good identification rates for signals from eight types of radar emitters across various signal-to-noise ratio environments, with an average feature extraction time of only 0.12 seconds per individual signal. The experiments confirm the effectiveness and real-time capabilities of the proposed method, indicating its significant engineering value.

Key words: specific emitter identification, feature extraction, wavelet basis, metric learning, signal representation