Reproduced code of HRAwCNN for automatic velocity analysis

This repository replicates a method of velocity analysis, but we do not provide the dataset because it is classified. Cite: Ferreira R S, Oliveira D A B, Semin D G, et al. Automatic velocity analysis using a hybrid regression approach with convolutional neural networks[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 59(5): 4464-4470.

Data preparation

You need prepara two segy(sgy) files which includes velocity spectra and CMP gather infomation, and a label file which includes the velocity labels. You have to build the h5 file for the index of samples, as shown in https://github.com/newbee-ML/MIFN-Velocity-Picking/blob/master/utils/BuiltStkDataSet.py

Implement

There are three parts for implementing the method proposed by Ferreira et al.:

1. Generate the CropNMO dataset for training Xception Network.
2. Training Xception Network.
3. Predict processing

Tips: You have to change a few path settings, if you want to test these method on your datasets.

Generate the CropNMO dataset for training Xception Network

python GenerateCNNData.py --dataset hade --CropSize 256,256
python GenerateCNNData.py --dataset dq8 --CropSize 256,256

Training Xception Network

python XceptionTrainMain.py

Training details:

• Velocity range [1000, 7000]
• NMO correction interval 50m/s, range [-1000, 1000]
• The shape of input NMO image is (256, 256)

Predict processing

python HRAwCNNPredMain.py

Prediction details:

• The shape of input NMO image is (256, 256)
• Time stride is 100 pixel

Test Results on two field datasets

DataSet	lrStart	optimizer	trainBS	VMAE
hade	0.0001	adam	32	173.5236
hade	0.0001	adam	16	190.6544
hade	0.001	adam	32	191.1893
hade	0.001	adam	16	225.2926
dq8	0.0001	adam	32	778.9616
dq8	0.001	adam	32	823.6368
dq8	0.0001	adam	16	887.0123
dq8	0.001	adam	16	1042.129