semi-blind-trace-deep-learning

This demo is a TensorFlow implementation of the algorithm explained in the paper “Semi-blind-trace algorithm for self-supervised attenuation of trace-wise noise” by Mohammad Mahdi Abedi, David Pardo, Tariq Alkhalifah. The accompanying data includes a small part of the 2004 BP velocity estimation benchmark model (with tripled trace interval) contaminated with trace-wise noise. The data and the code are provided without warranties. Please read the Terms of Use of the original data here: Link Please unzip the provided dataset, and run the code.
The semi_blind_trace.py code loads the data, defines the architecture, loss function, and custom training steps, trains the proposed model with the proposed algorithm, and plots the results. The “Results” folder includes a pretrained model, obtained using the provided code and data set.

For more information see the full paper: arxiv, wiley