Predicting the oil content of individual corn kernels combining NIR-HSI and multi-stage parameter optimization...

Area of Science:

• Agricultural Science
• Spectroscopy
• Machine Learning

Background:

• Predicting corn kernel oil content is crucial for breeding and processing.
• Traditional methods often lack speed, destructiveness, or rely on subjective parameter settings.
• Hyperspectral imaging (HSI) and machine learning (ML) offer non-destructive, rapid alternatives.

Purpose of the Study:

• To develop and validate an innovative multi-stage grid search technique for optimizing ML models in HSI-based corn oil content prediction.
• To overcome limitations of traditional parameter setting and improve prediction accuracy.
• To compare the performance of the proposed method against existing models like ACNNR.

Main Methods:

• Collected 270 corn kernel samples from diverse varieties and ear locations.
• Utilized hyperspectral imaging to capture spectral data.
• Implemented an automated model screening of 504 algorithm combinations followed by multi-stage grid search for parameter optimization.
• Evaluated models including SG+NONE+KS+PLSR, MA+LAR+Random+MLR, and ACNNR.

Main Results:

• Initial optimal models achieved R2 values of 0.8570 and 0.8523.
• Parameter optimization via multi-stage grid search improved R2 to 0.9045 and 0.8730 for the respective models.
• The ACNNR model achieved an R2 of 0.8878 and RMSE of 0.2243.
• The optimized multi-stage grid search method demonstrated superior prediction accuracy.

Conclusions:

• The developed multi-stage grid search technique significantly enhances the accuracy and adaptability of HSI-ML models for predicting corn kernel oil content.
• This method offers a robust and effective alternative to traditional approaches, suitable for field applications.
• The findings pave the way for more precise and efficient corn quality assessment.