[Study on Visual Identification of Corn Seeds Based on Hyperspectral Imaging Technology]

Area of Science:

• Agricultural Science
• Spectroscopy
• Image Processing

Background:

• Seed purity is critical for agricultural productivity and quality control.
• Traditional seed identification methods can be time-consuming and subjective.
• Advancements in hyperspectral imaging offer potential for objective and efficient seed analysis.

Purpose of the Study:

• To develop a visual identification method for corn seed cultivars using near-infrared hyperspectral imaging.
• To establish a robust model for accurate seed classification and purity assessment.
• To enable intuitive visualization of seed distribution in mixed samples for agricultural applications.

Main Methods:

• Acquisition of near-infrared hyperspectral images (874–1,734 nm) for 384 corn seed samples across 4 cultivars.
• Selection of 7 effective wavelengths (EWs) using the Successive Projection Algorithm (SPA).
• Development of a Partial Least Squares (PLS) model utilizing SPA-selected EWs for seed classification.

Main Results:

• The SPA-PLS model achieved high prediction accuracy, with Rc = 0.9177 and Rcv = 0.9115.
• Identification rates were 78.5% for the calibration set and 70.8% for the prediction set.
• Visual identification maps effectively displayed the distribution of different corn cultivars in mixture samples.

Conclusions:

• Near-infrared hyperspectral imaging combined with SPA-PLS is a viable method for visual corn seed identification.
• The developed technology provides an intuitive way to observe seed cultivar distribution in mixtures.
• This research supports efficient seed identification and screening in agricultural production.