ASABE Technical Library - Abstract

Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, view access and subscription options

Non-Destructive Quantitative Prediction of Moisture and Oil Content in Ground Coffee Based on FT-NIR Spectroscopy and Explainable AI

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2026 ASABE Annual International Meeting  2600092.(doi:10.13031/aim.202600092)
Authors:   Runyu Zheng, Mohammed Kamruzzaman
Keywords:   Fourier Transform near-infrared (FT-NIR) spectroscopy, explainable artificial intelligence (XAI), partial least squares regression (PLSR), wavelength selection, content determination.

Abstract. Rapid, non-destructive evaluation of chemical attributes in ground coffee is essential for quality control, process optimization, and product consistency. This study establishes a robust analytical framework for quantitative prediction of moisture and oil content using Fourier Transform near-infrared (FT-NIR) spectroscopy integrated with explainable artificial intelligence (XAI). FT-NIR spectra (1000-2500 nm) were collected from ground coffee samples and preprocessed using scatter correction, derivative transformation, and normalization techniques. Partial least squares regression (PLSR) was employed to develop predictive models, and wavelength selection was conducted using ten methods: regression coefficients (RC), variable importance in projection (VIP), successive projections algorithm (SPA), uninformative variable elimination (UVE), competitive adaptive reweighted sampling (CARS), genetic algorithm (GA), ant colony optimization (ACO), grey wolf optimizer (GWO), minimum redundancy maximum relevance (mRMR), and ReliefF. Results showed that GA produced the best model for moisture (R²c = 0.94, RMSEc = 0.11, RPDc = 4.05; R²p = 0.89, RMSEp = 0.17, RPDp = 2.96), while RC yielded the optimal model for oil (R²c = 0.92, RMSEc = 0.20, RPDc = 3.63; R²p = 0.83, RMSEp = 0.28, RPDp = 2.42). To enhance interpretability, SHapley Additive exPlanations (SHAP) identified key wavelengths associated with moisture (1006, 1209, 1344, 1435, 1534, 1717, 1751, and 2277 nm) and oil (1209, 1347, 1384, 1417, 1435, 1517, 1576, 1673, 1698, 1729, 1897, and 1965 nm) absorption features. The proposed FT-NIR-XAI workflow provides an efficient, reliable, and interpretable solution for assessing critical quality attributes in ground coffee, supporting real-time monitoring and intelligent decision-making in coffee production.

(Download PDF)    (Export to EndNotes)