A geophysically constrained 3D electromagnetic resistivity model database and its Deep Learning application on ATEM Inversion

A geophysically constrained 3D electromagnetic resistivity model database and its Deep Learning application on Airborne Transient Electromagnetic Inversion

This material includes a comprehensive 3D resistivity model database (3DRMD) and the code of fast 3D inversion operator guided by deep learning.

Input_traindata.npy  shape(14492, 24, 24, 23)  simples,dimension along X, dimension along Y, dimension along cut-off time(np.logspace(-5, -3, 21), plus transmitter coil and receiver coil Hight divide by 10.  The values are 21 base-10 logarithm of induced electromotive force values in T/s  with moment=1e4. The last two elements are the heights of the transmitting coil and the receiving coil (in meters) divided by 10. The entire input data will be further divided by 10 to reduce the magnitude's impact on the neural network.

Label_traindata.npy  shape(14492, 24, 24, 13)  simples,dimension along X, dimension along Y, dimension along ZThe values are 13 base-10 logarithm of resistivity

prediction.npy the same with Label_traindata.npy but a predicted resistivity from UNet.

To run prediction code, just runpython predict.py

Requirement:python 3.8tensorflow 2.10numpy 1.24