Seismic Synthetic Data Generator written in Odin

❗❗UNDER DEVELOPMENT❗❗

3D Synthetic seismic generator to build synthetic data, mainly for deep learning training. Mathematically define by combination of linear and non linear functions to create geological model of shearing and folding developed by Wu, et al., (2019) "Building realistic structure models to train convolutional neural networks for seismic structural interpretation"

The deep learning part mainly related to data and label generation. Label in this context is geological object which is fault. Faulting in this model defined by displacement field (Walsh and Watterson (1978); Georgsen et al., (2012) on Wu et al., (2019)

The paper by Wu et al., (2019) can be found here

Current Progress

• [✅] Generate horizontal layers

• [✅] Folding ($S_2$)

$$ Folding_{(S2)}(X, Y, Z) = a + \frac{1.5}{max(Z)} \times Z ( \sum_{k=1}^{k=N} b_k \times e^{\frac{(X-c_k)^2 + (Y-d_k)^2}{2\sigma^{2}_{k}}} ) $$

• input parameters guard

• Faulting

  • parameter parser

  • the funcitions and core logic

• Saving the final result

• some visualization for the sake of clarity of basically, everything in this repo

• documentation and proper module separation

Motivation

Why i'm making this? well simply i am in learning arc of low level language. I code extensively in Python, the loving filthy snake of slowness, jk, i love Python. I'm currenlty learning C-Zig-and-Odin, i kinda scared of C and found Zig slighlty complicated, then i bumped to Odin, which i found it, kinda less scary than those 2, and feels like Python if that makes sense.

All of the code were reproduced from this repo: python repo: 3D-Seismic-Image-Fault-Segmentation

How to use

1. open your terminal
2. clone and change dir to this repo

git clone https://github.com/kelreeeeey/OdinSeismicSyntheticGenerator && cd ./OdinSeismicSyntheticGenerator

3. make and run

   you can just run plain make or make opti to use -o:speed -no-bounds-check -disable-assert flag to build the binary

-> make opti && make run
odin build . -out:./bin/syndatagen.exe -o:speed -no-bounds-check -disable-assert
# odin build . -out:./bin/syndatagen.exe -o:aggressive -microarch:native -no-bounds-check -disable-assert
./bin/syndatagen.exe example_parameter.json
Reflection: 40000
Reflection[0]: 200
Done processing XY[0/40000] (0.0012511 seconds)
Done processing XY[8000/40000] (9.320025899999999 seconds)
Done processing XY[16000/40000] (18.811389999999999 seconds)
Done processing XY[24000/40000] (29.718454699999999 seconds)
Done processing XY[32000/40000] (41.0226983 seconds)
Gaussed-reflection 40000 [][]f64
Params XY[1000]: [PointXY{idx = 1000, x = 5, y = 0}], X[0]: [5], Y[0]: [0]
Layer Models: Reflection: len(40000), Layers: len(40000)
Layer Models: Reflection[0]: len(200), Layers[0]: len(200)
Creating things takes: 53.0258972 seconds

4. if you dont have make installed you can check makefile and choose in which way you want to build the binary

The executable binary will be found in ./bin subdir. This repo comes with example parameter in .json format. here some breakdown:

{
  "data_dim" : { // main parameters
    "patch_size": 128, // for final sampling
    "trace_size": 200, // dimension size in x-y-z direction.
    "layer_position": [ // geological layer of the model in depth/vertical/z direction
      0, 17, 30, 50, 88,
      100, 120, 130, 160, 199 ]
  },
  "folding" : { // folding parameter (see equation `Folding(S2)`)
    "a" : 5.0,
    // b,c,d and sigma should be the type of list.
    // b,c,d and sigma should have the same length, and should not
    // be empty. if lenghts do not match, will be aligned to the smallest one.
    "b" : [15.0, 5.2],
    "c" : [46.2, 36],
    "d" : [10,46],
    "sigma" : [14, 7.0]
  },
  "verbose": false
}

⚠️ if you want to make a new parameter, please copy the example and change it.

Reference

1. Building realistic structure models to train convolutional neural networks for seismic structural interpretation (Wu, et al., 2019)
2. Python project: 3D-Seismic-Image-Fault-Segmentation
3. Odin-Lang