Modal feature extraction utilizing image pyramid and dynamic mode decomposition

This paper presents a modal feature extraction method based on a multi-scale image pyramid and Dynamic Mode Decomposition (DMD) to address the inherent trade-off between accuracy and processing efficiency in structural modal analysis. By designing a nearest neighbor propagation compensation mechanism and an adjustable pyramid framework with intermediate scale layer fusion, combined with DMD, we achieve efficient and accurate multi-order modal parameter extraction. In simulations and experiments, the proposed method outperforms state-of-the-art pyramid methods in both identification accuracy and processing efficiency, particularly excelling in high-order modal feature extraction and noise robustness. The average modal confidence (MAC) improves by 6.5%, the root mean square error (RMSE) decreases by 48.8%, and the runtime is reduced by 8.1%. These results indicate that this study can provide a reliable and efficient technical method for long-term structural health monitoring.