Detecting Microcracks in Micrographs by a Neural Network

Figure 1. Micrograph of a rock fragment that resulted from drilling a well. Damage from drilling causes these microfractures and shape deformation (left) that normally correlate with rock composition (bottom right). These drilling-induced damages also correlate with geomechanical properties such as fracturability, elasticity, hardness, and stiffness (Young's modulus).

Objective: Use a FCNN or CNN to delineate the microfractures in the samples in the file here.

Procedure

  1. Install the GIMP picking software here. Go to the GIMP lab.
  2. Download the micrograph samples.
  3. Use a FCNN or CNN to train a network to detect the microfractures. You will have to identify the number of training pictures, and label them. You will use semantic segmentation, i.e. pixel classification, so you will use hand labels. Learn how to label with the GIMP lab.
  4. This can be divided among several people: one person uses U-Net CNN and the other uses Alexnet.

Report:

  1. PPT presentation should have the following slides: Title, Outline, Problem Definition and Proposed Solution, Procedure slides, Results slides, Conclusion, Future Work. Grab the EAGE London ppt at as an example.
  2. Write up should be a paper (e.g. 4-page extended abstract) that contains the following elements: Abstract, Intoduction, Procedures, Results, Conclusions. Perhaps 4-6 written pages, with as many figures as you need.