![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F16438831%2Fa78ff0da73185bd94aab913b4b62209d%2FDenosing.PNG?generation=1707973197689105&alt=media\")
Hi, Kagglers! š
\nEEG signals are easily influenced by external factors
\nExternal factors: Eye Movement, muscle movement
Drawbacks: Potential for data loss & Various types of noise with wide frequency bands
In Chris's How to Make Spectrogram form EEG Notebook
\nDmey: [cv: 0.77 -> cv: 0.82]
\nDb8: [cv: 0.77 -> cv: 0.80]
Alternative: Limiting the range to 0~20Hz using Butter low-pass filtering, followed by denoising within that range.
In Chris's WaveNet Starter Notebook
\nAfter Butter Low Pass Filtering
\nStratifiedGroupKFold: cv: 0.843
\nStratifiedGroupdKFold & db6: cv: 0.822
But also, There are also a lot of method to denoise
\n
šÆ I think if you want to make NN, it seems like good to build just simple CNN
\nIf you have any idea, feel free to tell me! Thanks for reading
", "messages": [ { "id": "2652915", "postDate": "02/15/2024 05:01:50", "content": "Hi, Kagglers! š
\nEEG signals are easily influenced by external factors
\nExternal factors: Eye Movement, muscle movement
Drawbacks: Potential for data loss & Various types of noise with wide frequency bands
In Chris's How to Make Spectrogram form EEG Notebook
\nDmey: [cv: 0.77 -> cv: 0.82]
\nDb8: [cv: 0.77 -> cv: 0.80]
Alternative: Limiting the range to 0~20Hz using Butter low-pass filtering, followed by denoising within that range.
In Chris's WaveNet Starter Notebook
\nAfter Butter Low Pass Filtering
\nStratifiedGroupKFold: cv: 0.843
\nStratifiedGroupdKFold & db6: cv: 0.822
But also, There are also a lot of method to denoise
\n
šÆ I think if you want to make NN, it seems like good to build just simple CNN
\nIf you have any idea, feel free to tell me! Thanks for reading
", "rawMarkdown": "Hi, Kagglers! š\n\n###BackGroud\nEEG signals are easily influenced by external factors\n\n**External factors**: `Eye Movement`, `muscle movement`\n\n###Method1: Wavelet Transfomer\n`Drawbacks`: Potential for data loss & Various types of noise with wide frequency bands\n\nIn Chris's How to Make Spectrogram form EEG Notebook\nDmey: **[cv: 0.77 -> cv: 0.82]**\nDb8: **[cv: 0.77 -> cv: 0.80]**\n\n`Alternative`: Limiting the range to 0~20Hz using Butter low-pass filtering, followed by denoising within that range.\n\nIn Chris's WaveNet Starter Notebook\nAfter Butter Low Pass Filtering\nStratifiedGroupKFold: **cv: 0.843**\nStratifiedGroupdKFold & db6: **cv: 0.822**\n\n###Method2: Neural Network\nBut also, There are also a lot of method to denoise\n\n\n\nšÆ **I think if you want to make NN, it seems like good to build just simple CNN**\n\nIf you have any idea, feel free to tell me! Thanks for reading", "votes": null }, { "id": "2654118", "postDate": "02/15/2024 21:50:04", "content": "what do you think of these articles:
\nA multi-artifact EEG denoising by frequency-based
\ndeep learning
\nhttps://arxiv.org/pdf/2310.17335.pdf
\nA novel convolutional neural network model to remove
\nmuscle artifacts from EEG
\nhttps://arxiv.org/pdf/2010.11709.pdf
They seem to use : https://github.com/ncclabsustech/EEGdenoiseNet to dataset to purely just denoise to pure EEGs:
\ndo you think there are ways to use this for our use case?
First of all, Thanks for sharing good Document!
\nIf you want to use EEGdenoiseNet, you need pure EEGs signal.
\nBut as you know, in kaggle dataset there's no pure eeg signal.
So, you have to make pure eegs with filtering or transforming.
\nšÆ But, in my opinion it's better just use filtering and transforming not CNN Models
\nšÆ and in my experiement, it was good to use Wavlet after filtering!
", "rawMarkdown": "First of all, Thanks for sharing good Document! \nIf you want to use EEGdenoiseNet, you need pure EEGs signal. \nBut as you know, in kaggle dataset there's no pure eeg signal. \n\nSo, you have to make **pure eegs with filtering or transforming.**\n\nšÆ But, in my opinion it's better just use filtering and transforming not CNN Models\n\nšÆ and in my experiement, it was good to use Wavlet after filtering!", "votes": null } ], "comments": [ { "id": 2654118, "author_name": "tashin47", "author_url": "", "post_date": "02/15/2024 21:50:04", "content": "what do you think of these articles:
\nA multi-artifact EEG denoising by frequency-based
\ndeep learning
\nhttps://arxiv.org/pdf/2310.17335.pdf
\nA novel convolutional neural network model to remove
\nmuscle artifacts from EEG
\nhttps://arxiv.org/pdf/2010.11709.pdf
They seem to use : https://github.com/ncclabsustech/EEGdenoiseNet to dataset to purely just denoise to pure EEGs:
\ndo you think there are ways to use this for our use case?
First of all, Thanks for sharing good Document!
\nIf you want to use EEGdenoiseNet, you need pure EEGs signal.
\nBut as you know, in kaggle dataset there's no pure eeg signal.
So, you have to make pure eegs with filtering or transforming.
\nšÆ But, in my opinion it's better just use filtering and transforming not CNN Models
\nšÆ and in my experiement, it was good to use Wavlet after filtering!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2652915": "Hi, Kagglers! š\n\n###BackGroud\nEEG signals are easily influenced by external factors\n\n**External factors**: `Eye Movement`, `muscle movement`\n\n###Method1: Wavelet Transfomer\n`Drawbacks`: Potential for data loss & Various types of noise with wide frequency bands\n\nIn Chris's How to Make Spectrogram form EEG Notebook\nDmey: **[cv: 0.77 -> cv: 0.82]**\nDb8: **[cv: 0.77 -> cv: 0.80]**\n\n`Alternative`: Limiting the range to 0~20Hz using Butter low-pass filtering, followed by denoising within that range.\n\nIn Chris's WaveNet Starter Notebook\nAfter Butter Low Pass Filtering\nStratifiedGroupKFold: **cv: 0.843**\nStratifiedGroupdKFold & db6: **cv: 0.822**\n\n###Method2: Neural Network\nBut also, There are also a lot of method to denoise\n\n\n\nšÆ **I think if you want to make NN, it seems like good to build just simple CNN**\n\nIf you have any idea, feel free to tell me! Thanks for reading", "2654118": "what do you think of these articles:\nA multi-artifact EEG denoising by frequency-based\ndeep learning \nhttps://arxiv.org/pdf/2310.17335.pdf\nA novel convolutional neural network model to remove\nmuscle artifacts from EEG\nhttps://arxiv.org/pdf/2010.11709.pdf\n\nThey seem to use : https://github.com/ncclabsustech/EEGdenoiseNet to dataset to purely just denoise to pure EEGs:\ndo you think there are ways to use this for our use case?", "2654347": "First of all, Thanks for sharing good Document! \nIf you want to use EEGdenoiseNet, you need pure EEGs signal. \nBut as you know, in kaggle dataset there's no pure eeg signal. \n\nSo, you have to make **pure eegs with filtering or transforming.**\n\nšÆ But, in my opinion it's better just use filtering and transforming not CNN Models\n\nšÆ and in my experiement, it was good to use Wavlet after filtering!" }, "source": "meta" }