SCIENCE@home

PLSELM: A lightweight modeling approach for low-data calibration in near-infrared spectroscopy.

Xiaqiong Fan,Lijin Shang,Shuo Zhao,Jixing Fan,Senlin Zhang,Qiong Yang,C. Wu,Yulin Liu,Tiejun Yang,Hongchao Ji

Published 2025 in Analytica Chimica Acta

ABSTRACT

BACKGROUND Near infrared (NIR) spectroscopy is widely used as a rapid analytical technique in various fields for its advantages of on-line monitoring and non-destructive testing. It can provide rich chemical information and is of great significance for studying the structure, composition and changes of substances. Reliable calibration remains a major challenge in near-infrared (NIR) spectroscopy, especially under low-data conditions or across instruments with varying configurations. To address this, we propose PLSELM, a lightweight modeling calibration method, which combines Partial Least Squares (PLS) score matrices and Ensemble Extreme Learning Machine (ELM). RESULTS To address this, we propose PLSELM, a lightweight modeling calibration method, which combines Partial Least Squares (PLS) score matrices and Ensemble Extreme Learning Machine (ELM). By modeling the relationship between latent PLS features and concentration values, PLSELM provides a fast, robust, and transferable calibration framework. To evaluating the performance, five diverse NIR spectral data, including 21 sets of concentration indicators from 10 different spectrometers, were used for benchmarking comparison. These NIR spectra have different wavelength ranges, resolutions, lengths, and a wide range of concentrations. Results demonstrate that PLSELM has excellent calibration performance, outperforming conventional PLS, Support Vector Regression, and deep learning-based models. PLSELM also has great suitability in low-data learning and calibration transfer analysis. In addition, PLSELM model has good robustness, which is manifested in that it is not sensitive to the randomness of sample division and the randomness of hidden layer nodes. PLSELM only took 0.5 s to finished the PLSELM and PLS models on corn data. SIGNIFICANCE The comprehensive comparison results indicate that the PLSELM method is a robust NIR calibration method, which performs well in various spectral wavelength ranges, resolutions, lengths, and a wide range of concentrations. In summary, PLSELM offers a practical and scalable solution for NIR calibration, with excellent potential for use in real-world analytical applications involving limited data or heterogeneous instruments.

PUBLICATION RECORD

CITATION MAP

EXTRACTION MAP

CLAIMS

  • No claims are published for this paper.

CONCEPTS

  • No concepts are published for this paper.

REFERENCES

Showing 1-34 of 34 references · Page 1 of 1

CITED BY