Seismic data interpretation
All, I am working on interpreting seismic data for some purpose. Basically these come as 3D volumes of data, one for each seismic attribute. Now I have training samples with this data, and I am using a classifier to classify the output as 4 classes.
Inputying these readings parallels and merging the result.
The issue is my loss doesn't seem to decrease after few epochs.
Any suggestions pls. Used BatchNorm after Conv3D, maxpooling3D, repeated and then dropout and finally flatten.
Thanks in advance.
Well, thanks for your reply.
I am now getting a good convergence for Training data, accuracy going upto 100% almost. Doing binary classification now (High / Low producing Wells). However, cannot get to increase accuracy (F1 score) of Test Data above 85%. Using keras functional API to input 17 seismic attributes (after normalization), pass through Conv2D, MaxPooling and merge the dense layers , before doing classification. Using AdaDelta, binary_crossentropy. Works good, but there is a overfitting with training data. Already have Dropout of 0.5 after two layers of Conv2D, followed by MaxPooling, but increasing drop-out (or adding another dropout after first Conv2D) is not helping. Any suggestions welcome.