Full waveform inversion (FWI) is a powerful tool to reconstruct material distributions with damages, such as voids or cracks, based on sparsely measured wave propagation data. It is used in infrastructure health monitoring to identify the damage location and severity. However, the process is often ill-posed and computationally expensive. Alternatively, neural networks (NNs) can be trained by supervised learning based on data from domains with known defects [1] to predict damage. However, such a purely supervised approach can lead to wrong predictions for signals outside the training data set. A more reliable alternative is to use NNs embedded within classical FWI, where Neural Networks are used as the parametrization of the material field [2], utilizing the adjoint formulation commonly used in FWI. This combination of NN with FWI (NN-based FWI) produces more physically meaningful local minima than classical FWI by reducing noise and improving the accuracy of the recovered void shape. NN-based FWI can be further improved by combining it with transfer learning. The proposed technique utilizes supervised pretraining to provide a better NN weight initialization, leading to faster convergence of FWI. The network is pretrained to predict the unknown material field using the gradient information from the first iteration of conventional FWI. Transfer learning is, thus, used to provide an initial guess for the subsequent NN-based FWI. We demonstrate that using gradient information in transfer-learning-based FWI also offers reliable predictions for data outside the training set and reduces the number of iterations required for convergence [3]. Even in cases where the initial guess is not accurate, transfer learning FWI can still recover the damaged shape but requires more training epochs. References 1. J. Rao, Y. Fangshu, M. Huadong, S. Kollmannsberger, und E. Rank. „Quantitative reconstruction of defects in multi-layered bonded composites using fully convolutional network-based ultrasonic inversion“. Journal of Sound and Vibration 542 (2023): 117418. https://doi.org/10.1016/j.jsv.2022.117418. 2. L. Herrmann, T. Bürchner, F. Dietrich, and S. Kollmannsberger, “On the use of neural networks for full waveform inversion,” Comput. Methods Appl. Mech. Eng., vol. 415, p. 116278, Oct. 2023, doi: 10.1016/j.cma.2023.116278. 3. S. Kollmannsberger, D. Singh, and L. Herrmann, “Transfer Learning Enhanced Full Waveform Inversion,” in 2023 IEEE/ASME Internationa