Classification of Frozen Corn Seeds Using Hyperspectral VIS/NIR Reflectence Imaging

Area of Science:

• Agricultural Science
• Spectroscopy
• Image Processing

Background:

• Freeze damage significantly impacts corn seed viability and crop yield.
• Accurate and efficient methods for assessing seed quality are crucial for agriculture.
• Hyperspectral imaging offers a non-destructive approach for seed analysis.

Purpose of the Study:

• To classify three degrees of freeze damage in corn seeds using VIS/NIR hyperspectral imaging.
• To compare various image preprocessing, wavelength selection, and classification methods.
• To evaluate two visualization methods (M2M and M2P) for freeze-damage representation.

Main Methods:

• VIS/NIR hyperspectral imaging of corn seed embryos.
• Application and comparison of four preprocessing methods (MSC, SNV, 5-3 smoothing).
• Evaluation of four wavelength selection algorithms (SPA, PCA).
• Comparison of three classification models (PLS-DA, KNN, SVM).
• Analysis of two visualization techniques (M2M, M2P).

Main Results:

• The combination of 5-3 smoothing and SPA wavelength selection improved signal-to-noise ratio and achieved over 90% classification accuracy.
• The Mean Spectrum to Pixel Spectrum (M2P) method provided better classification results than Mean Spectrum to Mean Spectrum (M2M).
• M2P visualization resulted in fewer misclassified samples and better representation of freeze damage.

Conclusions:

• VIS/NIR hyperspectral imaging combined with optimized preprocessing (5-3 smoothing) and wavelength selection (SPA) is effective for classifying freeze-damaged corn seeds.
• The M2P visualization technique enhances the ability to accurately assess and visualize freeze damage in corn seed embryos.
• This approach offers a promising non-destructive method for quality control in corn seed production.