$A^2DM$ : Enhancing EEG artifact removal by fusing artifact representation into the time-frequency domain
This is the Official PyTorch implementation of our Cognitive Computation paper $A^2DM$ : Enhancing EEG artifact removal by fusing artifact representation into the time-frequency domain .
In our model, we first construct a pretraining dataset to generate artifact representations. We use the code from https://github.com/ncclabsustech/EEGdenoiseNet to generate EEG signal segments with artifacts of different signal-to-noise ratios and artifact types. Labels are created based on the corresponding artifact data, and a classification task is used to train the model (the input is EEG segments with artifacts, and the output is the artifact representation). The network structure is in the file Classifier_network_base.py.
We recommend retraining the artifact representation module according to your specific task.
The data and model files are available in the Releases section and can be accessed directly.
Similarly, we use the code from https://github.com/ncclabsustech/EEGdenoiseNet to generate the artifact-removed dataset for training the model (the input is EEG segments with artifacts, and the output is the artifact-removed EEG segments).
We provide the test set used in our study (test_input.npy and test_output.npy) ，the pre-trained model (basemodel.pth), and the trained model(a2dm.pkl)
You can directly run the test using:
python eval_all.py