Discrimination of New and Aged Seeds Based on On-Line Near-Infrared Spectroscopy Technology Combined with Machine...

Area of Science:

• Agricultural Science
• Spectroscopy
• Chemometrics

Background:

• Maize seed harvest year significantly impacts seed vitality and crop yield.
• Accurate identification of maize seed age is essential for agricultural management.
• Near-infrared (NIR) spectroscopy offers a non-destructive method for seed analysis.

Purpose of the Study:

• To design and validate an on-line near-infrared (NIR) spectra collection system for distinguishing maize seeds based on harvest year.
• To compare the performance of different chemometric models for classifying maize seed harvest years.
• To optimize spectral data preprocessing and feature selection for improved classification accuracy.

Main Methods:

• Development of an on-line NIR spectra collection device (899-1715 nm).
• Evaluation of chemometric models including Partial Least Squares Discriminant Analysis (PLS-DA), LS-SVM, KNN, and ELM.
• Application of spectral preprocessing techniques (SGS, SNV, MSC, SG-D1, SG-D2, Norm) and variable selection methods (MC-UVE, CARS, BOSS, SPA).

Main Results:

• The PLS-DA model demonstrated optimal recognition performance for maize seed harvest years compared to other tested models.
• Spectral preprocessing and variable selection significantly improved classification accuracy.
• The optimized MC-UVE-BOSS-PLS-DA model, using Norm preprocessed data and 93 features, achieved a classification accuracy of 88.75%.

Conclusions:

• The self-designed NIR collection system is a viable tool for identifying maize seed harvest years.
• Combining NIR spectroscopy with advanced chemometric methods enables accurate and non-destructive seed age classification.
• This technology can support better agricultural practices by ensuring the use of appropriate aged maize seeds.