Development of a rapid detection method for maize seed purity using a modular high-throughput near-infrared non-destructive testing system

Abstract

The identification of maize seed varieties is crucial for ensuring high agricultural production quality, enhancing food security, regulating the seed market, advancing technological development, and supporting environmental sustainability. This study applied a high-throughput near-infrared detection system to identify various maize seed varieties. Using spectral data, different preprocessing methods and classification models were explored to construct a classification model for mixed-coated maize seeds (nine classifications). Results indicated that full-spectrum technology effectively distinguishes between mixed-coated varieties. Optimization with CARS and SPA algorithms identified the optimal model, SG-CARS-SPA-LR, which used 60 feature bands to achieve classification accuracies of 0.87 and 0.86 for calibration and test sets, respectively. Additionally, a three-class model for single-coated seeds was developed, enhancing the classifier’s generalization ability through feature selection and model optimization. Findings demonstrate that full-spectrum technology performs best for identifying varieties with the same coating type, especially uncoated seeds. The SG-CARS-SPA-LR model achieved exceptional classification accuracy (0.99) for both calibration and test sets using only 15 feature bands, underscoring this method’s efficiency and suitability for high-throughput seed analysis applications.