Hyperspectral imaging combined with texture features for maize hybrid purity detection: a multi-model comparison...

Area of Science:

• Agricultural Science
• Biotechnology
• Spectroscopy

Background:

• Maize hybrid seed purity is vital for optimal crop performance.
• Distinguishing hybrid seeds from self-pollinated ones is challenging due to similar phenotypes.
• Conventional visual inspection is insufficient for accurate purity assessment.

Purpose of the Study:

• To develop a non-destructive and interpretable framework for detecting maize hybrid seed purity.
• To integrate hyperspectral imaging and RGB-derived texture features for enhanced accuracy.
• To achieve high-throughput purity detection meeting national standards.

Main Methods:

• Acquisition of hyperspectral images from embryo and endosperm sides of maize seeds.
• Construction and comparison of texture-based, spectral, and texture-spectral fusion models.
• Application of wavelength selection algorithms (CARS, SPA, Sync2D) and machine learning classifiers (SVM, PLS-DA).

Main Results:

• Embryo-side spectral information proved more stable and discriminative than endosperm-side.
• Texture-spectral fusion models achieved high accuracies (0.99-1.00), exceeding the 97% national requirement.
• Optimal models reduced spectral variables by 88%-95%, identifying key features like mean saturation and specific spectral regions (450-462 nm).

Conclusions:

• The developed framework offers a high-throughput, low-dimensional, and interpretable solution for maize hybrid seed purity detection.
• Integration of hyperspectral imaging and texture analysis significantly improves purity assessment accuracy.
• The method ensures genetic purity, contributing to improved maize breeding and agricultural productivity.