UnseenSignalTFG: a signal-level expansion method for unseen acoustic data based on transfer learning

This study introduces a transfer learning-based approach for signal-level expansion of unseen acoustic signal data, aiming to address the scarcity of acoustic signal data in a specific domain. By establishing connections and sharing knowledge between the source and target domains, the method successfully mitigates cross-domain disparities, overcoming challenges posed by unavailable data in the target domain, thereby elevating the quality and precision of data expansion. Diverging from conventional methods that predominantly emphasize feature-level expansion, the proposed approach accentuates the preservation of data signal integrity and effectively achieves the expansion of unseen class samples within the target domain.The effectiveness of this method has been validated across four different types of signal datasets. In the bearing dataset, the expansion of unseen data achieved accuracies of 99% at the signal level and 95% at the spectral level. These experimental results not only demonstrate the method’s advantages in augmenting both seen and unseen data but also highlight its effectiveness and application potential in achieving comprehensive expansion of target signals.