[Purity Detection Model Update of Maize Seeds Based on Active Learning].

Area of Science:

• Agricultural Science
• Spectroscopy
• Machine Learning

Context:

• Seed purity is crucial for agricultural quality control.
• Hyperspectral imaging offers non-destructive seed analysis.
• Model generalization is challenged by variations in harvest origin and year.

Purpose:

• To investigate the effectiveness of active learning for updating hyperspectral imaging-based seed purity detection models.
• To compare active learning with random selection and Kennard-Stone algorithms for model updating.
• To enhance the accuracy and stability of maize seed purity detection models.

Summary:

• Active learning algorithms were employed to incorporate representative samples into existing hyperspectral imaging models for maize seed purity detection.
• Models updated with active learning demonstrated substantial improvements in prediction accuracy for new samples from different harvest years.
• Active learning outperformed random selection and Kennard-Stone methods in enhancing model performance and generalization.

Impact:

• Enables rapid and accurate updates of seed purity detection models.
• Improves the reliability of non-destructive seed quality assessment.
• Facilitates more robust agricultural product quality control systems.