coronalyze.core.pca

Principal Component Analysis (PCA/KLIP) for PSF subtraction.

Implements the efficient ‘Snapshot Method’ for SVD on large image cubes, particularly suited for coronagraphy (where pixels >> frames). All functions are JIT-compilable and fully differentiable.

Functions

get_pca_basis(ref_cube, n_modes)	Compute PCA basis vectors from a reference cube (Snapshot Method).
pca_subtract(science_image, basis, mean_ref)	Project science image onto PCA basis and subtract the model.

Module Contents

coronalyze.core.pca.get_pca_basis(ref_cube, n_modes)

Compute PCA basis vectors from a reference cube (Snapshot Method).

Uses eigendecomposition of the NxN covariance matrix instead of full SVD, which is O(N²xP) vs O(P²xN) for pixels P >> frames N.

Args:

ref_cube: Reference image cube of shape (N_frames, Height, Width). n_modes: Number of principal components (modes) to keep.

Returns:

basis: The top n_modes eigen-images, shape (n_modes, Height*Width). mean_ref: The mean of the reference cube, shape (Height, Width).

Parameters:

• ref_cube (jax.numpy.ndarray)

• n_modes (int)

Return type:

tuple[jax.numpy.ndarray, jax.numpy.ndarray]

coronalyze.core.pca.pca_subtract(science_image, basis, mean_ref)

Project science image onto PCA basis and subtract the model.

Args:

science_image: 2D science image, shape (Height, Width). basis: PCA basis from get_pca_basis, shape (n_modes, Height*Width). mean_ref: Mean reference from get_pca_basis, shape (Height, Width).

Returns:

Residual image with PSF model subtracted, shape (Height, Width).

Parameters:

• science_image (jax.numpy.ndarray)

• basis (jax.numpy.ndarray)

• mean_ref (jax.numpy.ndarray)

Return type:

jax.numpy.ndarray