Two-Pass

📜 Acquisition Protocol

The Two-Pass algorithm is designed to accelerate hyperspectral acquisition, traditionally performed point-by-point (e.g., snake scan, raster scan).

The acquisition process is as follows:

1. A fast initial acquisition (at low SNR) is performed for each point in the imaged area.

2. The data undergoes two processing paths:

   • Spatial Compression:
     • Dimension reduction is applied along the spectral axis to generate a single, high-contrast image of the sample.
     • The resulting image is segmented into super-pixels.
     • For each super-pixel, a representative (centroid) is selected, effectively performing spatial compression.
   • Spectral Compression:
     • Each spectrum from the hyperspectral image is decomposed in the Fourier domain using the DFT.
     • The first harmonics are extracted to identify the essential pixels, corresponding to spectral compression.

3. The subset of centroid pixels is unmixed relative to the essential pixels to produce a concentration matrix.

4. The essential pixels are rescanned at a slower pace (with high SNR).

Finally, the super-pixel map, the concentration matrix of centroids, and the high-SNR essential pixels are recombined to reconstruct the hyperspectral image.

---

📊 Results

Using this protocol, the following results can be obtained (the "UnSCR" column corresponds to Two-Pass):

Spatial Reconstruction

Spectral Reconstruction

Acquisition Time Comparison

---

🚀 Installation

Clone the repository and navigate to the project directory:

conda create -n twopass python=3.9 -y
conda activate twopass
git clone https://github.com/ValentinGilet/Two-Pass.git
cd Two-Pass
pip install -r requirements.txt

---

📚 Reference

V. Gilet et al., “Superpixels meet essential spectra for fast Raman hyperspectral microimaging,” Opt. Express, vol. 32, no. 1, p. 932, Jan. 2024, doi: 10.1364/OE.509736.