spectral-process 0.2.1

Python tools for spectral data preprocessing

spectral-process

Python tools for spectral data preprocessing — baseline correction, scatter correction, normalization, and derivative filtering.

---

Overview

spectral-process is a python library for applying preprocessing techniques on the NIR, MIR, hyperspectral, and other spectral dataset. It provides a set of well-tested implementations of the most widely used data preprocessing techniques, which can be applied to chemometrics and agricultural spectroscopic datasets.

---

Installation

From GitHub (recommended for now):

git clone https://github.com/mah-iasri/spectral-process.git
cd spectral-process
pip install -e .

From PyPI (coming soon):

pip install spectral-process

---

Usage Examples

Import the package

from spectral_process import (
    baseline_correct,
    apply_savitzky_golay_derivative,
    snv_transformation,
    unit_vector_normalization,
    multiplicative_scatter_correction,
)
import numpy as np

---

1. ALS-based Baseline Correction

Removes fluorescence background or sloping baselines common in the spectra.

# Simulate 10 spectra with 200 wavelength points
spectra = np.random.rand(10, 200) + np.linspace(0, 1, 200)

# Apply ALS baseline correction
corrected = baseline_correct(spectra, lam=1e5, p=0.01, niter=10)

# Also retrieve the estimated baselines
corrected, baselines = baseline_correct(
    spectra, lam=1e5, p=0.01, niter=10, return_baselines=True
)

Parameters to tune:

• lam — smoothness of baseline (typical range: 1e4 to 1e7)
• p — asymmetry (typical range: 0.001 to 0.1)

---

2. Savitzky-Golay Derivative Filtering

Smooths spectra and computes derivatives to remove baseline offsets and enhance spectral features.

# First derivative with 21-point window, 2nd order polynomial
first_derivative = apply_savitzky_golay_derivative(
    spectra,
    window_length=21,
    polyorder=2,
    deriv=1
)

# Second derivative
second_derivative = apply_savitzky_golay_derivative(
    spectra,
    window_length=21,
    polyorder=2,
    deriv=2
)

---

3. Standard Normal Variate (SNV)

Corrects for multiplicative scatter effects and baseline shifts on a per-sample basis. Each spectrum will have mean = 0 and std = 1.

snv_spectra = snv_transformation(spectra)

# Using sample standard deviation (ddof=1)
snv_spectra = snv_transformation(spectra, ddof=1)

---

4. Unit Vector Normalization

Scales each spectrum to unit L2 norm. Useful before applying machine learning models.

normalized = unit_vector_normalization(spectra)

# Raise an error if any zero-norm spectra are found
normalized = unit_vector_normalization(spectra, handle_zero='error')

---

5. Multiplicative Scatter Correction (MSC)

Corrects for light scattering differences between samples. Uses the mean spectrum as reference by default.

msc_spectra = multiplicative_scatter_correction(spectra)

# Use a custom reference spectrum (e.g. from a calibration set)
reference = np.mean(spectra[:50], axis=0)
msc_spectra, ref_used = multiplicative_scatter_correction(
    spectra,
    reference=reference,
    return_reference=True
)

---

Full Preprocessing Pipeline Example

import numpy as np
from spectral_prep import (
    baseline_correct,
    apply_savitzky_golay_derivative,
    snv_transformation,
)

# Load your spectral data (n_samples × n_wavelengths)
spectra = np.load("nir_spectra.npy")   # shape: (100, 1050)

# Step 1: Baseline correction
spectra_bc = baseline_correct(spectra, lam=1e5, p=0.01, niter=10)

# Step 2: Savitzky-Golay first derivative
spectra_sg = apply_savitzky_golay_derivative(
    spectra_bc, window_length=21, polyorder=2, deriv=1
)

# Step 3: SNV normalization
spectra_ready = snv_transformation(spectra_sg)

print(spectra_ready.shape)  # (100, 1050)

---

API Reference

Baseline Correction

Function	Description	Key Parameters
als(y, lam, p, niter)	ALS baseline correction for a single spectrum	lam: smoothness; p: asymmetry
baseline_correct(spectra, lam, p, niter, return_baselines)	ALS correction for multiple spectra	return_baselines=True to get baselines

Derivative Filtering

Function	Description	Key Parameters
apply_savitzky_golay_derivative(spectra, window_length, polyorder, deriv, delta, mode)	SG filter / derivative	deriv=0 for smoothing only; deriv=1 or 2 for derivatives

Normalization

Function	Description	Key Parameters
snv_transformation(spectra, ddof)	Standard Normal Variate	ddof=0 (population) or 1 (sample)
unit_vector_normalization(spectra, handle_zero)	L2 unit vector normalization	handle_zero: 'warn', 'ignore', 'error'

Scatter Correction

Function	Description	Key Parameters
multiplicative_scatter_correction(spectra, reference, return_reference, eps)	MSC scatter correction	reference: custom 1D reference; return_reference=True to retrieve it

---

Input Requirements

All functions expect:

• Input: 2D numpy.ndarray of shape (n_samples, n_features)
• Output: 2D numpy.ndarray of the same shape
• Invalid inputs raise a ValueError with a descriptive message

---

Dependencies

Package	Version
numpy	≥ 1.23
scipy	≥ 1.10

---

Authors

Name	Affiliation	Contact
Md Ashraful Haque	ICAR-IASRI, New Delhi, India	[e-mail]
Avijit Ghosh	ICAR-IGFRI, Jhansi, India	[e-mail]

Citation

If you use spectral-process in your research, please cite:

@software{haque2026spectralprocess,
  author    = {Haque, Md Ashraful and 
                Ghosh, Avijit},
  title     = {spectral-process: Python tools for spectral data preprocessing},
  year      = {2026},
  publisher = {GitHub},
  url       = {https://github.com/mah-iasri/spectral-process}
}

Or in plain text:

Haque, M.A. & Ghosh, A. (2026). spectral-process: Python tools for spectral data preprocessing. GitHub. https://github.com/mah-iasri/spectral-process

---

License

This project is licensed under the MIT License — see the LICENSE file for details.

---

Contributing

Contributions are welcome! Please open an issue first to discuss any proposed changes. See CONTRIBUTING.md for guidelines.