Predicting potato dry matter content by incorporating spectral statistical features

• A triple-optimization framework was developed for prediction of potato DMC. • Spectral statistical features were proposed to capture local band variation. • Data fusion of spectral and statistical features improved model accuracy. • The SDSS–CARS–PLSR model achieved R²P = 0.80 and RPD > 2.0 for DMC prediction. Accurate and high-throughput prediction of crop quality parameters is essential for intelligent agricultural production and automated grading. Dry matter content (DMC) is a key indicator of potato quality and processing performance. To address the limitations of hyperspectral modeling in capturing sample heterogeneity and localized spectral features, this study presents a multi-source modeling approach by integrating near-infrared hyperspectral reflectance (780-2000 nm) with spectral statistical features, Spectral Dispersion (SD) and Spectral Smoothness (SS). A dataset of 442 potato samples was analyzed using a triple optimization strategy, which includes sample subset optimization, key wavelength selection, and statistical feature fusion. The resulting models demonstrated robust predictive performance, with coefficients of determination for the prediction ( R 2 P ) of 0.80 and 0.75, root mean square errors of prediction (RMSEP) of 1.32% and 1.44%, and ratio of performance to deviation (RPD) values > 2. These results confirm the effectiveness of the proposed fusion strategy for DMC prediction, providing a scalable and transferable approach for rapid quality assessment of crop biochemical properties and supports intelligent quality assessment in precision agriculture.