SRResNet: A Super-Resolution Residual Network for seismic signal denoising based on PyTorch

Copyright (c) 2023 Zhengjie Zhang ([email protected])

We would like to thank Dr. H. Wang for her inspiration and help in this work.

• This is the core code of the project Traffic seismic data denoising based on machine learning.
• We do not share the data for other purposes.
• Note that the use of UNet.py in this package still needs to be optimized, and we are still debugging and modifying it.

Network Architecture

Installation

Via Anaconda (recommended):

conda create -n SRResNet python=3.8
conda activate SRResNet
conda install numpy==1.23.4 matplotlib==3.6.3 obspy==1.3.0 torch==1.10.1 scikit-learn==1.2.2 torchsummary==1.5.1 pandas==1.5.3

Clone source codes

set your working directory at /data/

cd /data/
git clone https://github.com/zhangzj1209/SRResNet.git
unzip SRResNet.zip
cd SRResNet/

Description

Please create several folders in path /data/SRResNet/

mkdir -r data           ! used to store training data and validation data
mkdir -r label          ! used to store training label and validation label
mkdir -r save           ! used to store train model and predict result
mkdir -r predict_data   ! used to store prediction data
mkdir -r predict_label  ! used to store prediction label

• If you want to use this network to do your work, please modify the contents of My_Dataset in dataset.py.
• The number of residual block layers of the network can also be modified in line 29 of SRResNet.py.