I have a stack of signals that I'd like to perform an ML algorithm to chop them up into regions determined by their characteristics. The full signal looks a bit like this:
And a fragment of it looks like this:
In particular, I'd like to highlight the bits of the trace where the discontinuities are.
All of the traces I have are likely to be of varying lengths, and with variable amounts of noise on them so some of the discontinuities are hard to detect (e.g. around the 5000 mark on this trace). I do have a set of reasonably well labelled data that I would like to use to train some form of ML to do this to try and better regular numerical techniques. I've so far done some basic stuff with a regular neural net, and a CNN, on a sliding window through the data which doesn't do too badly but I'm keen to explore other options. In particular, is there a 1-D variant on RCNN/YOLO algorithms that could be used to return a set of windows that bound each discontinuity? it should be noted that the discontinuities may be of varying width, too
Asked
Active
Viewed 65 times
1
