Any starter sources which can be referred to for starting with using digital signal processing techniques on these images?
", "messages": [ { "id": 1301520, "postDate": "2021-05-11T05:53:25.490Z", "content": "Any starter sources which can be referred to for starting with using digital signal processing techniques on these images?
", "rawMarkdown": "Any starter sources which can be referred to for starting with using digital signal processing techniques on these images?", "votes": 9 }, { "id": 1310059, "postDate": "2021-05-16T13:18:31.390Z", "content": "Like you, I'm also interested in exploring whether any signal processing techniques could be useful in pre-processing or augmenting the data set. I think some of the more interesting techniques available involve understanding more about the dimensions of the spectrogram data provided. I've asked for some clarifications here:
\nhttps://www.kaggle.com/c/seti-breakthrough-listen/discussion/239339
An upvote may help improve and garner a response from the organizers. Short of getting a direct answer, I've also started reading through the papers posted on this discussion thread:
\nhttps://www.kaggle.com/c/seti-breakthrough-listen/discussion/239245
From the papers, it might be possible to back into more detail on the GBT spectrograms although since they've taken all this data and performed some additional processing and trimming while synthesizing some needles it can't be fully assumed that data descriptors in the papers fully align with the data provided for this competition.
\nThis may be a long way of saying that until more detail is known about the spectrograms, they're little more than simple images at this point which doesn't help enable a lot of signal processing and instead lends itself to the CNN approaches you already see on the leaderboard. There are a few data augmentations you could potentially leverage from a couple of the papers in the second link but I'm presently hoping for either feedback on my first thread or to discover a useful nugget in the published literature to help me bridge the gap.
", "rawMarkdown": "Like you, I'm also interested in exploring whether any signal processing techniques could be useful in pre-processing or augmenting the data set. I think some of the more interesting techniques available involve understanding more about the dimensions of the spectrogram data provided. I've asked for some clarifications here: \n[https://www.kaggle.com/c/seti-breakthrough-listen/discussion/239339](https://www.kaggle.com/c/seti-breakthrough-listen/discussion/239339)\n\nAn upvote may help improve and garner a response from the organizers. Short of getting a direct answer, I've also started reading through the papers posted on this discussion thread:\n[https://www.kaggle.com/c/seti-breakthrough-listen/discussion/239245](https://www.kaggle.com/c/seti-breakthrough-listen/discussion/239245)\n\nFrom the papers, it might be possible to back into more detail on the GBT spectrograms although since they've taken all this data and performed some additional processing and trimming while synthesizing some needles it can't be fully assumed that data descriptors in the papers fully align with the data provided for this competition.\n\nThis may be a long way of saying that until more detail is known about the spectrograms, they're little more than simple images at this point which doesn't help enable a lot of signal processing and instead lends itself to the CNN approaches you already see on the leaderboard. There are a few data augmentations you could potentially leverage from a couple of the papers in the second link but I'm presently hoping for either feedback on my first thread or to discover a useful nugget in the published literature to help me bridge the gap.", "votes": 3, "replies": [ { "id": 1318156, "postDate": "2021-05-22T04:28:18.887Z", "content": "Thanks for the help, it was certainly helpful going through the links and the other discussions.
", "rawMarkdown": "Thanks for the help, it was certainly helpful going through the links and the other discussions." } ] }, { "id": 1301609, "postDate": "2021-05-11T06:56:18.750Z", "content": "https://www.youtube.com/watch?v=T1_F1Kn2vu0
", "rawMarkdown": "https://www.youtube.com/watch?v=T1_F1Kn2vu0", "votes": 1, "replies": [ { "id": 1301639, "postDate": "2021-05-11T07:12:50.557Z", "content": "Thanks, that's certainly helpful!
", "rawMarkdown": "Thanks, that's certainly helpful!" } ] } ], "comments": [ { "id": 1310059, "author_name": "Jeffrey Egan", "author_url": "", "post_date": "2021-05-16T13:18:31.390000", "content": "Like you, I'm also interested in exploring whether any signal processing techniques could be useful in pre-processing or augmenting the data set. I think some of the more interesting techniques available involve understanding more about the dimensions of the spectrogram data provided. I've asked for some clarifications here:
\nhttps://www.kaggle.com/c/seti-breakthrough-listen/discussion/239339
An upvote may help improve and garner a response from the organizers. Short of getting a direct answer, I've also started reading through the papers posted on this discussion thread:
\nhttps://www.kaggle.com/c/seti-breakthrough-listen/discussion/239245
From the papers, it might be possible to back into more detail on the GBT spectrograms although since they've taken all this data and performed some additional processing and trimming while synthesizing some needles it can't be fully assumed that data descriptors in the papers fully align with the data provided for this competition.
\nThis may be a long way of saying that until more detail is known about the spectrograms, they're little more than simple images at this point which doesn't help enable a lot of signal processing and instead lends itself to the CNN approaches you already see on the leaderboard. There are a few data augmentations you could potentially leverage from a couple of the papers in the second link but I'm presently hoping for either feedback on my first thread or to discover a useful nugget in the published literature to help me bridge the gap.
", "votes": 3, "replies": [ { "id": 1318156, "author_name": "Suryansu Dash", "author_url": "", "post_date": "2021-05-22T04:28:18.887000", "content": "Thanks for the help, it was certainly helpful going through the links and the other discussions.
", "votes": 0, "replies": [] } ] }, { "id": 1301609, "author_name": "Rishabh Chaurasia", "author_url": "", "post_date": "2021-05-11T06:56:18.750000", "content": "https://www.youtube.com/watch?v=T1_F1Kn2vu0
", "votes": 1, "replies": [ { "id": 1301639, "author_name": "Suryansu Dash", "author_url": "", "post_date": "2021-05-11T07:12:50.557000", "content": "Thanks, that's certainly helpful!
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1301520": "Any starter sources which can be referred to for starting with using digital signal processing techniques on these images?", "1310059": "Like you, I'm also interested in exploring whether any signal processing techniques could be useful in pre-processing or augmenting the data set. I think some of the more interesting techniques available involve understanding more about the dimensions of the spectrogram data provided. I've asked for some clarifications here: \n[https://www.kaggle.com/c/seti-breakthrough-listen/discussion/239339](https://www.kaggle.com/c/seti-breakthrough-listen/discussion/239339)\n\nAn upvote may help improve and garner a response from the organizers. Short of getting a direct answer, I've also started reading through the papers posted on this discussion thread:\n[https://www.kaggle.com/c/seti-breakthrough-listen/discussion/239245](https://www.kaggle.com/c/seti-breakthrough-listen/discussion/239245)\n\nFrom the papers, it might be possible to back into more detail on the GBT spectrograms although since they've taken all this data and performed some additional processing and trimming while synthesizing some needles it can't be fully assumed that data descriptors in the papers fully align with the data provided for this competition.\n\nThis may be a long way of saying that until more detail is known about the spectrograms, they're little more than simple images at this point which doesn't help enable a lot of signal processing and instead lends itself to the CNN approaches you already see on the leaderboard. There are a few data augmentations you could potentially leverage from a couple of the papers in the second link but I'm presently hoping for either feedback on my first thread or to discover a useful nugget in the published literature to help me bridge the gap.", "1301609": "https://www.youtube.com/watch?v=T1_F1Kn2vu0" } }