# DL for geological features (structures) mapping – Problem Definition

## What is the problem?

### Informal description
I need a program that will recognize and outline geological features in photographs.
* The levels:
* Mapping the fractures (binary)
* Mapping and classifying the features

### Formal description
> A computer program is said to learn from experience _E_ with respect to some class of tasks _T_ and performance measure _P_, if its performance at tasks in _T_, as measured by _P_, improves with experience _E_. **Tom Mitchell**

- **Task (T)**: Map geological features (structures)
- **Experience (E)**: A dataset comprising images and corresponding binary masks
- **Performance** **(P**): Classification accuracy, Intersection over Union (IoU), and Area under the Curve (AuC), Dice, ...

### ==Assumptions==

* While fractures are generally small-scale, shallow, planar cracks in rocks (Barton & Zoback, 1992; Bonnet et al., 2001; Segall & Pollard, 1983), faults span a broad range of length scales ($10^{−6} – 10^3$ km) and surface-to-depth widths ($1–10^2$ km), and have a complex 3D architecture (Mattéo et al. 2021).
* Fault traces appear as dark lineaments in images, forming dense networks (Mattéo et al. 2021).
* Fault anf fracture traces have complex, curvilinear shapes and assemble, connect or intersect in an even more complex manner, yet partly deterministic (Mattéo et al. 2021).
* GIS environments allow fault attributes such as trace hierarchical importance, thickness, interruptions, connections, and slip mode to be labeled in various ways (various line thicknesses, colors, symbols, etc.) (Mattéo et al. 2021)
* While the fractures are short open cracks with no relative motion, the faults are longer features with clear evidence of lateral slip (en echelon segmentation, pull-apart connections, slickensides) (Mattéo et al. 2021).
* The geological structures have:
* intrinsic variables such as **geometry**, **soft linkage** and **segmentation over multiple scales**; [@thiele2017]
* extrinsic variables such as **natural variations in colour**, **shadows**, **glare** and/or **incomplete geological exposure**. [@thiele2017]

<!--  -->

* Features definitions:
* Fractures are "a change in contrast" that is sow subtle that need some experience to get it right
* scale dependant (observations scale), mechanics (...)
* ...
* The elevation matter to the model (if available)
* Spatial dependency?
* Noise?
* Geological features are defined by?: morphology, location, timing and orientation [@thiele2017]
* Vertical and horizontal data ??
*

### Similar problems
* Medical Image Segmentation (Vessels(retinal,...)...)
* Building Surface Crack Detection
* Road extraction from remote sensing images
* Plants veins extraction
* Industrial Quality Control (defects detection)

=> highlight limitations
=> ! conceptual drift (concept is it changing over time) = not really
=> get algorithms and data transformations that could be adopted

## Why does the problem need to be solved?

### Motivation
Research, PhD, =)
- The explosion in the amounts of high-resolution data acquired from remote sensors (Satellite, airborne, UAVs) [@thiele2017]
- Geological structures hold valuable information about the history of the Earth, earthquake mechanics, and are critical for seeking resources such as water, petroleum, and minerals.
- mineral exploration, CO2 sequestration, groundwater, and geothermal energy.
- Monitor changes
- risk, drilling, ...

### Solution Benefits
* Automatic: No human input or very minimal
* Fast: Save time significantly compared to manual and semi-automatic annotation processes, where annotating a small area could require hours
* Accurate: Satisfying results
* Scale: Support for multiple scales
* Extract the full geological value of high-quality datasets [@thiele2017]

### Solution Use
* How: get the model parameters, fine-tune to current dataset, predict ...
* Lifetime: Until another intelligent breakthrough supersedes it.
* Functional and nonfunctional requirements for the solution:
* > ?

## How to solve the problem?

#### How to solve the problem manually?
- Digitizing by hand.
- As a programming exercise?
* Vanilla computer vision??
* Edge, line detectors
*

### Data collection and pre-processing
* [[dl-frac-map-data]]
> * Prototypes and experiments to perform (there are a gold mine because they will highlight questions and uncertainties about the domain that could be explored).

#### Update the previous sections

### Model ?!
* [[dl-frac-map-model]]

### Post-processing
Binary image not very useful,
* Vectorization (from raster)
* Width
* Intersections
* Connections ...

### Literature

- Chudasama, B., Ovaskainen, N., Tamminen, J., Nordbäck, N., Engström, J., & Aaltonen, I. (2024). Automated mapping of bedrock-fracture traces from UAV-acquired images using U-Net convolutional neural networks. Computers & Geosciences, 182, 105463. https://doi.org/10.1016/j.cageo.2023.105463
- Mattéo, L., Manighetti, I., Tarabalka, Y., Gaucel, J.-M., van den Ende, M., Mercier, A., Tasar, O., Girard, N., Leclerc, F., Giampetro, T., Dominguez, S., & Malavieille, J. (2021). Automatic Fault Mapping in Remote Optical Images and Topographic Data With Deep Learning. Journal of Geophysical Research: Solid Earth, 126(4), e2020JB021269. https://doi.org/10.1029/2020JB021269
- [ ] Thiele, S. T., Grose, L., Cui, T., Cruden, A. R., & Micklethwaite, S. (2019). Extraction of high-resolution structural orientations from digital data: A Bayesian approach. Journal of Structural Geology, 122, 106–115. https://doi.org/10.1016/j.jsg.2019.03.001

- ...

---

Problem definition: [Ref](https://machinelearningmastery.com/how-to-define-your-machine-learning-problem/), [Ref2](https://machinelearningmastery.com/machine-learning-checklist/)