GAN_II

The reconstruction of images on optical wavelengths with high precision has always been challenging for astronomers. Here, we present the application of a Generative Adversarial Network (GAN) for Intensity Interferometry and perform image reconstruction for a fast-rotator.

The Python files follow the information of the model of the star and baseline configuration. There is need to run these scripts one by one in given order below...

To create N number of Images....

Run generate_image.py

The parameters can be changed using these two scripts

1. ellipse.py: Define a function to create N number of 2-dimensional (x, y) grids to draw a contour map of an ellipse with different parameters (radius, inclination, position angle, axis ratio)
2. generate_image.py: The parameters of the ellipse are defined here and saved in .jpg format

To simulate the Telescope and Baselines Configuration.......

Run generate_baseline.py

The parameters can be changed using these four scripts

1. aperture.py: It defines the shape of the aperture for each telescope and visualizes it on the observational plane. It also defines a function to create an array of baseline' names and length on
   observational plane.
2. obstime.py: It calculates the number of steps at each observational night due to observational intervals. It also calculates the start and end time after each observational interval. There is another definition to calculate an array of all Julian days for all given observational intervals.
3. obspoint.py: Since everything is calculated in radians. There are definitions to convert degrees and hours to radians. There is also another function that calculates the change in baseline due to earth rotation
4. generate_baseline.py: It generates the variational baseline with time, which is used to simulate the II signal for any source and telescope array. However, we need to define some parameters here, exa, (1) Array of starting times of the observational day. (2) Array of ending times for the observational day (3) calculate the number of steps for the given observational interval (average time), so the array of Julian time (4) Define the telescope's position and name. Then get the baseline's name and length. (5) Define the position of of source in the sky coordinate. (6) Calculate the variational baseline according to the Earth's rotation. (7) Plot the covered (u, v) plane with baseline in a .png file for visualization (8) Convert the observational plane to grayscale, re-arrange pixels, and save the baseline in .npy format

Create Input data for GAN

Run generate_pspec.py

The parameters can be changed using these two scripts

1. image_noise.py: It defines the to add salt and paper noise to the ellipsoid image so that it can create noisy II data for the input to GAN. It also defines the function to add two images. Here, one image is a stellar source (fast-rotator) and the other is an II signal using this source and generated baselines.
2. generate_pspec.py: It calculates the power spectrum (the absolute value of the Fourier transform) of each noised image using salt and paper noise def in "image_noise.py". It also calculates the visibility for a given baseline (power spectrum X baseline). Later, it normalizes the signal and adds both ground truth (ellipsoid image) and signal (visibility) side by side, which works as an input image for GAN. At the end, it saves the images in the train, test, and validation folders.

Check the loss function

Run generate_loss.py for the visualization of loss function for two different parameters. The loss is stored in "logs.pkl" file.

Train the model with GAN for the given signal

Run main_def_test.py to check the definition. Run gan_def_test.py to train the model.

The parameters can be changed using these three scripts

1. main_def.py: This script prepares the set of definitions used for the GAN structure while training.
2. gan_def, py: This script describes the GAN network as a generator and a Discriminator. Using these networks, GAN can train the model.
3. gan_def_test.py: This script test the "gan_def.py" and model is trained here.

Visualization of trained result

Run generate_reconst.py

Statistical Visualization of Result

Run generate_moment.py

The definition of moments is defined in "moment.py"

Enjoy!!!