The Signal Feature Extraction of Graph Fourier Transform on the Constructed Graph
Junxiong Wang, Xiang Pan · 2021
Motivated by graph Fourier transform (GFT), we propose a novel signal processing method for feature extraction. Mapping the original signal into a transformed space by looking for a suitable basis, thereby observing the new feature of the signal in the transform domain, has always been the research content of signal processing. In a finite-dimensional transformation space, the original signal can be expressed as a vector. By constructing a set of orthogonal vectors containing the original signal, they become the eigenvectors of the Laplacian matrix of the constructed graph. In this way, the original signal is mapped to a point after the GFT, and the signal feature is obtained in the transformation domain. In the underwater target recognition experiment, the proposed method is suitable for underwater acoustic signal processing and has better performance than several common data construction graph model methods.