oveys96 / eit_seminar

Comparison of different image reconstruction methods with different injection pattern

Comparison of numerical 2D reconstruction methods and the corresponding requirements and impact of parameters using the pyEIT package.

Install Requirements

All requirements are provided inside the requirements.txt.

pip3 install -r requirements.txt # Linux, macOS, Windows
pip install -r requirements.txt  # Windows

Run the examples

From the example folder, pick one demo and run!

You will see the results of (2D) forward and inverse computing with different parameter in each demo.

Note: the following images may be outdated due to that the parameters of a EIT algorithm may be changed in different versions of pyEIT.

(2D) forward and inverse computing

Using codes/examples/one_obj_cent.ipynb

REMARK:

Three different image reconstruction methods are shown.
The number of electrodes in each row is equal.
The image reconstruction method in each colomn is same.

Result:

Jacobian is the slowest and Backprojection is the fastest method.
GREIT is the trade off between Jacobian and Backprojection for time and resolution.

Using codes/examples/one_obj_side.ipynb

REMARK:

Three different image reconstruction methods are shown.
The number of electrodes in each row is equal.
The image reconstruction method in each colomn is same.

Result:

Jacobian is the slowest and Backprojection is the fastest method.
GREIT is the trade off between Jacobian and Backprojection for time and resolution.

Using codes/examples/two_obj_sides.py

REMARK:

Three different image reconstruction methods are shown.
The number of electrodes in each row is equal.
The image reconstruction method in each colomn is same.

Result:

Jacobian is the slowest and Backprojection is the fastest method.
GREIT is the trade off between Jacobian and Backprojection for time and resolution.
All three methods are good with goal of the detection of the number of the separated objects.

Using codes/examples/twoD_obj_sides.py

REMARK:

Three different image reconstruction methods are shown.
The number of electrodes in each row is equal.
The image reconstruction method in each colomn is same.
The target is trying to detect two objects with same conductivity but different size.

Result:

Jacobian is the slowest and Backprojection is the fastest method.
GREIT is the trade off between Jacobian and Backprojection for time and resolution.

Using codes/examples/adjandopp.py

REMARK:

Three different image reconstruction methods are shown.
The number of electrodes in any case is equal.
The image reconstruction method in each colomn is same.
The target is trying to compare adjacent with opposite injection pattern.

Result:

Opposite injection pattern showes separated objects clearer.
GREIT is the best and Backprojection is the worst.

Using codes/examples/geometry.py

REMARK:

Three different image reconstruction methods are shown.
The number of electrodes in any case is equal.
The image reconstruction method in each colomn is same.
The goal is trying to detect the object like triangle with sharp edges with adjacent and opposite injection pattern.

Result:

GREIT almost fails even in separating the objects.
Jacobian has the highest resolution.
Adjacent injection pattern causes better results for objects with sharp edges.
Simulating the edges is difficult task and almost all these methods fail in it. We should use other tools for post processing the image.

Contact

Email: [email protected]

eit_seminar's Issues

Improvements

• Add short markdown descriptions of what we see in the plotted images. For example, say which method is "better" (faster/resolution properties, etc.).
• Please add "titles" to the plots which have adjacent and opposite injection patterns in the adjandopp.ipynb notebook.
• Use only one color map. Decide on your own if you want to use cmap=plt.cm.viridis or cmap=plt.cm.RdBu.
• You can also add a README.md file or a wiki to the repository, where all the images are listed one below the other with their parameter description and properties. The description here can be identical to that in the respective notebooks.