Adaptive conditional diffusion-based data augmentation for NIR spectroscopy quantitative analysis under limited labeled samples

Peng Li

doi:10.1016/j.saa.2026.128647

Adaptive conditional diffusion-based data augmentation for NIR spectroscopy quantitative analysis under limited labeled samples

2026-08-23

0

PRE

AI

DOI：10.1016/j.saa.2026.128647

Original

Original request for help

Save

Abstract

En 中文

• Proposes an Adaptive Conditional Diffusion Model (ACDM) for generating NIR spectra under continuous regression labels. • Introduces a hybrid conditioning mechanism combining classifier-free guidance and Feature-wise Linear Modulation (FiLM) to enhance sensitivity to label variations. • Develops a label-guided adaptive noise injection network (CGN) to improve controllability and generation accuracy. • Incorporates Squeeze-and-Excitation (SE) modules for spectral feature recalibration, enhancing structural fidelity of generated spectra. • Demonstrates superior regression performance under limited labeled samples across multiple datasets, including corn kernels, American ginseng, and pharmaceutical tablets.

Adaptive conditional diffusion-based data augmentation for NIR spectroscopy quantitative analysis under limited labeled samples

Abstract

Journal

Organization

Cited Papers