Estimation Model for Maize Multi-Components Based on Hyperspectral Data.

Area of Science:

• Agricultural Science
• Analytical Chemistry
• Spectroscopy

Background:

• Corn seed quality is crucial for agriculture and relies on accurate component analysis.
• Traditional methods for assessing corn seed components are often time-consuming and destructive.
• Non-invasive techniques are needed for rapid and efficient quality evaluation.

Purpose of the Study:

• To develop a non-invasive method for simultaneously detecting multiple key components in corn seeds.
• To optimize a chemometric model for predicting water, fat, protein, and starch content.
• To evaluate the effectiveness of hyperspectral imaging combined with advanced algorithms for corn seed analysis.

Main Methods:

• Hyperspectral imaging (1100-2498 nm) was used to collect data from corn seed embryos.
• Image segmentation identified the germ region of interest.
• Spectral data preprocessing (Detrending Transformation - DT) and characteristic wavelength extraction (Competitive Adaptive Reweighted Sampling - CARS) were performed.
• Least Squares Support Vector Machine (LSSVM) models, optimized with Particle Swarm Optimization (PSO), were constructed for component prediction.

Main Results:

• The Detrending Transformation (DT) algorithm was optimal for spectral preprocessing.
• The CARS algorithm effectively extracted characteristic wavelengths, reducing data redundancy.
• The optimized CARS-PSO-LSSVM model achieved high prediction accuracy for water, fat, protein, and starch content (R values: 0.9884, 0.9490, 0.9864, 0.9687).

Conclusions:

• Hyperspectral imaging integrated with the DT-CARS-PSO-LSSVM algorithm provides an effective non-destructive method for corn seed quality assessment.
• This approach offers a scientific basis for evaluating corn seed quality and advances non-destructive testing technologies.
• The developed method enables rapid and accurate detection of key components, supporting agricultural applications.