Sugarbeet Seed Germination Prediction Using Hyperspectral Imaging Information Fusion

Area of Science:

• Agricultural Science
• Plant Science
• Spectroscopy

Background:

• Seed germination rate is crucial for agricultural productivity and quality control.
• Traditional germination testing is time-consuming and destructive.
• Developing rapid, nondestructive methods for germination prediction is essential.

Purpose of the Study:

• To develop and validate a nondestructive method for predicting sugarbeet seed germination using hyperspectral imaging (HIS).
• To analyze spectral and image features for improved germination prediction accuracy.

Main Methods:

• Single sugarbeet seeds were analyzed using HIS for nondestructive image segmentation.
• Spectral data underwent pretreatment (SNV+1D) and characteristic wavelengths were identified using Kullback-Leibler (KL) divergence.
• Image features were extracted using the gray-level co-occurrence matrix (GLCM).
• Partial least squares discriminant analysis (PLS-DA), CatBoost, and support vector machine radial-basis function (SVM-RBF) models were employed using spectral, image, and fusion features.

Main Results:

• Fusion features demonstrated superior predictive performance compared to spectral or image features alone.
• The CatBoost model, utilizing fusion features, achieved the highest prediction accuracy of 93.52%.
• Extracted characteristic wavelengths and image features were validated for their relevance.

Conclusions:

• Hyperspectral imaging combined with fusion features provides an accurate and nondestructive approach for predicting sugarbeet seed germination.
• This method holds potential for enhancing seed selection and quality assessment processes in agriculture.