With the continuous advancement of modern technology, more types of radar and related technologies are continuously being developed, and the identification of radar emitter signals has gradually become a very important research field. This paper focuses on the identification of modulation types in radar emitter signal identification. We propose a weighted normalized Singular-Value Decomposition (SVD) feature extraction algorithm, which is based on the perspective of data energy and SVD. The filtering effect of complex SVD is analyzed, as well as the influence of the number of rows of data matrix on the decomposition results, and the recognition effect of different classification models. The experimental results show that the algorithm has better filtering and recognition effects on common radar signals. Under –20 dB, the cosine similarity values of the reconstructed and original signals remain at about 0.94, and the recognition accuracy remains above 97% under a confidence level of 0.65. In addition, experiments show that the weighted normalized SVD feature extraction algorithm has better robustness than the traditional Principal Component Analysis (PCA) algorithm.