Corn seed dataset based on hyperspectral and RGB images

Area of Science:

• Agricultural Science
• Computer Vision
• Data Science

Background:

• Accurate maize seed characterization is crucial for agricultural productivity.
• Multimodal imaging offers enhanced data for phenotypic analysis.
• Developing robust datasets is essential for machine learning in agriculture.

Purpose of the Study:

• To create a comprehensive dataset of maize seeds using hyperspectral and RGB imaging.
• To facilitate research in seed variety classification and phenotypic analysis.
• To support advancements in precision agriculture and machine learning.

Main Methods:

• Utilized an HY-6010-S hyperspectral imaging system (400-1000 nm) and an RGB industrial camera.
• Acquired multimodal data from approximately 2400 maize seed samples across 12 varieties.
• Applied preprocessing techniques including noise reduction, background removal, and band selection using HHIT software and Python.

Main Results:

• Generated a large-scale multimodal dataset simulating phenotypic analysis of maize seeds.
• The dataset encompasses spectral and visual information for detailed seed characterization.
• Data processing ensured accuracy and reliability for downstream applications.

Conclusions:

• The developed dataset is valuable for seed variety classification and phenotypic analysis.
• This resource will advance machine learning applications in precision agriculture.
• Multimodal imaging provides a powerful approach for agricultural data acquisition.