Sparse Time-frequency Transform via Deep Learning and Transfer Learning: Part II-Transfer Learning and Field Data Application

Based on the STF-Unet model proposed in the first part of this study, we introduce a transfer learning (TL) model to enhance the generalization performance of the STF-Unet model, which is called the STFUnet with the transfer learning (STF-UTL) model. First, we use a small amount of field data to generate the training labels, i.e. the sparse TF spectra. Then, we utilize an adaptive TL method to fine-tune the pre-trained STF-Unet model. After validating the fine-tuned STF-UTL model, we apply it to a 3D field data volume to delineate seismic geological structures. Moreover, the detailed comparisons with the traditional STFT are introduced to demonstrate the validity and effectiveness of the proposed STF-UTL model for characterizing the fluvial channels.