A Cost-Effective and Scalable Machine Learning Approach for Quality Assessment of Fresh Maize Kernel Using NIR...

Area of Science:

• Agricultural Science
• Spectroscopy
• Machine Learning

Background:

• Developing accurate near-infrared (NIR) calibration models for fresh maize quality traits is time-consuming and costly.
• Traditional methods require large sample sets, posing challenges in breeding programs.

Purpose of the Study:

• To introduce a novel machine learning approach, the Prediction-Correction Neural Network (PCNN), for effective NIR calibration model development.
• To enable accurate quality trait prediction using small sample sets augmented with synthetic data.

Main Methods:

• Utilized a Prediction-Correction Neural Network (PCNN) model with NIR spectroscopy data.
• Employed small calibration sample sets augmented with synthetically generated data.
• Compared PCNN performance against Partial Least Squares (PLS) and Artificial Neural Networks (ANN).

Main Results:

• PCNN achieved high accuracy (RPD 2.821-4.862, R² 0.869-0.951) for traits like amylopectin and protein using only 32 samples.
• Accurate predictions for sugars (fructose, glucose, sucrose) were obtained with 62 samples (RPD >2, R² ≥0.747).
• PCNN outperformed PLS (38.99%-63.20% RPD improvement) and ANN (7.07%-25.82% RPD improvement).

Conclusions:

• PCNN offers a highly efficient, accurate, scalable, and cost-effective solution for rapid quality evaluation in fresh maize.
• The method is applicable to NIR model development for small sample sets in various crops, including forage maize, rice, wheat, and barley.
• PCNN significantly enhances NIR calibration model development, addressing traditional limitations in time, cost, and labor.