WSPSorter: a real-time integrated system for non-destructive purity sorting of watermelon seeds based on rotary singulation and near infrared spectroscopy.

Background: Watermelon seed purity is a critical determinant of varietal authenticity and production quality. In hybrid seed production, however, distinguishing between maternal seeds and hybrid seeds at the single-seed level remains challenging because their phenotypic differences are subtle. Conventional purity identification methods are often destructive, labor-intensive, and time-consuming and are poorly suited to the practical demand for rapid, continuous, single-seed online sorting. Methods: To address these challenges, this study developed the Watermelon Seed Purity Sorter (WSPSorter), an integrated real-time sorting system based on rotary singulation, near-infrared (NIR) spectroscopy, and pneumatic actuation. The system achieves stable single-seed delivery through a rotary seed-metering mechanism, acquires single-seed spectral information in real time, and combines NIR spectroscopy with an Extreme Gradient Boosting (XGBoost) classification model to discriminate maternal seeds from hybrid seeds. Results: The optimal model, combining wavelet denoising, the successive projections algorithm (SPA), and XGBoost, achieved accuracies of 98.89% on the validation set and 97.78% on the independent test set. WSPSorter achieved an average inter-sample interval of 254.88 ms, corresponding to a practical throughput of approximately 4.0 seeds per second. Sorting performance, evaluated against PCR-derived reference labels, reached 94.04% for maternal seeds and 90.05% for hybrid seeds. Discussion: These results indicate that WSPSorter effectively connects single-seed spectral sensing with real-time physical sorting, providing a feasible engineering solution for rapid, non-destructive, and automated purity sorting of watermelon seeds.