Single-shot Hyperspectral Imaging via Deep Chromatic Aberration Deconvolution

Code repository for manuscript titled:

Single-shot Hyperspectral Imaging via Deep Chromatic Aberration Deconvolution

Corey Zheng, Austin Barton, Mohammad Taher, Abdulaziz Memesh

Abstract: This work addresses the challenge of chromatic aberration in snapshot hyperspectral imaging, where the introduction of a third spectral dimension complicates encoding information into a two-dimensional detector plane. While traditional methods rely on scan-based approaches, our proposed method aims to enhance the quality of hyperspectral images by mitigating distortions inherent in snapshot acquisitions by leveraging a blind deconvolution approach with a U-Net neural network architecture for single-shot hyperspectral imaging. Our approach allows real-time correction without the need for complex scanning mechanisms nor knowledge of point spread functions. Through experimental validation, we demonstrate the efficacy of our method in preserving image structure and spectral composition information, contributing to improved imaging throughput and simplified hardware requirements. Our best performing model is capable of restoring spectral information among test data, even in pixel locations with highly varying wavelength intensities, while deblurring and restoring an approximation of the latent sharp image. Our paper represents a simple yet effective approach to circumventing issues in snapshot hyperspectral imaging, providing a practical solution for applications in medical imaging, agriculture, materials identification, and geological surveillance.

Paper

Setup

Option One (Recommended):

• Open Anaconda Prompt
• Navigate to src. Then, use the following command to create the environment: conda env create -f environment.yaml
• Activate the environment with: conda activate hyperspec
• Verify installation with: conda list

Option Two:

• Open Command Prompt and run the following command in the repository directory: pip install -r requirements.txt
• This also works for virtual environments but isn't recommended
• Verify installation with: pip list

Directory

src

Folder contents overview:

• Data synthesization.
• Baseline method (RL Deconvolution).
• Data loading.
• Model creation (U-Nets).
• Training and evaluation.
• Figure generation.
• Pixel comparison.

Folders and files summary:

• data: Data stored as h5 files.
• DatasetFormation: Data synthesis programs.
• loss_grapher: Loss graphing script that interacts with loss txt files generated by models and each model's folder for storing loss plots.
• main.py: Initialize hyperparameters and begin training and/or evaluation. Specify paths for loading and saving.
• model: The model creation, model summary script, and folders for model summary tables.
• RL_deconvolution: Richardson-Lucy deconvolution script for baseline comparison.
• saved_models (not remotely published): The folder to save models into. This will be done locally on your own device.
• Utils: Folder that contains utility functions for loading the data, choosing device, importing packages, etc.

overleaf

• Manuscript TeX files.

figs

• Highlighted figures including:
  • System process flowchart.
  • Data synthesis.
  • Point spread functions.
  • Simulated optical system.
  • Best results.
  • Degenerate results.
  • Loss curves.
  • Model architecture.
  • Metric tables.

Acknowledgements

This project is done as part of CS 7643 Deep Learning at Georgia Tech. We would like to directly thank Professor Danfei Xu and all of the teaching faculty for CS 7643 Deep Learning, Fall 2023 at Georgia Tech.