search

smousavi05 / EQTransformer

EQTransformer, a python package for earthquake signal detection and phase picking using AI.
https://rebrand.ly/EQT-documentations
MIT License
305 stars 148 forks source link

False negatives #62

Closed Cthuulhaa closed 3 years ago

Cthuulhaa commented 3 years ago

Hi,

do you see any false negative on these stations? It is a mining explosion (ML 1.1), with data available for 6 stations, but the event was detected on only one. Most of the traces look noisy, but I think I do "see something" on some of them (e.g. on station WIMM), but I am not an expert at detecting events...

Cheers!

ge falks ge flt1 sx muhb sx neub sx wimm th bonn

smousavi05 commented 3 years ago

@Cthuulhaa Hi I do see a weak signal on FLT1. The one on WIMM can be an event as well but this station look like to be a short period one, is that correct? that might be a reason for not picking it up. You may want to try the other model available here or/and try the prediction module with uncertainty estimation on.

Cthuulhaa commented 3 years ago

Hi! Yes, these are all HH* channels (with a sampling rate of 100). I have used EqT_model.h5. What is the difference between both models?

smousavi05 commented 3 years ago

the second model has less false negatives but with a cost of possible false positives. So you need to use higher threshold values to reduce false positives if you got many.