Solar multi-object multi-frame blind deconvolution with a spatially variant convolution neural emulator
The study of astronomical phenomena through ground-based observations is always challenged by the distorting effects of Earth’s atmosphere. Traditional methods of post-facto image correction, essential for correcting these distortions, often rely on simplifying assumptions that limit their effectiveness, particularly in the presence of spatially variant atmospheric turbulence. Such cases are often solved by partitioning the field-of-view into small patches, deconvolving each patch independently, and merging all patches together. This approach is often inefficient and can produce artifacts. Recent advancements in computational techniques and the advent of deep learning offer new pathways to address these limita- tions. This paper introduces a novel framework leveraging a deep neural network to emulate spatially variant convolutions, offering a breakthrough in the efficiency and accuracy of astronomical image deconvolution. By training on a dataset of images convolved with spatially invariant point spread functions and validating its general- izability to spatially variant conditions, this approach presents a significant advancement over traditional methods. The convolution emulator is used as a forward model in a multi-object multi-frame blind deconvolution algorithm for solar images. The emulator enables the deconvolution of solar observations across large fields of view without resorting to patch-wise mosaicking, thus avoiding artifacts associated with such techniques. This method represents a significant computational advantage, reducing processing times by orders of magnitude.
The emulator can be trained within the emulator directory. The script train_cond.py trains a conditional
U-Net to carry out the spatially variant convolution. Just edit the file to modify the config dictionary
that points to the location of the training set. Feel free to modify the rest of training parameters.
The trained weights can be downloaded here.
Using the emulator trained in the paper, you can deconvolve solar images. An example of how to proceed with
your observations is found in the examply.py script on the momfbd directory.
First download the Stable ImageNet-1K Stable ImageNet-1K database.
The directory database contains all the machinery to produce the training dataset. A large file containing all JPEG images reshaped
to the desired training size can be generated with reshape_imagenet_stablediff.py, while the final
training set can be generated with db_imagenet_stablediff.py. This file convolves all images with random
point spread functions in a certain range of Fried parameters and for certain configurations of telescopes. Feel
free to modify them according to your use case.