Soil data augmentation and model construction based on spectral difference and content difference.

Area of Science:

• Agricultural Science
• Analytical Chemistry
• Data Science

Background:

• Spectroscopy-based chemometrics for soil analysis are limited by small sample sizes.
• Developing precision agriculture and monitoring land quality requires robust analytical models.

Purpose of the Study:

• To propose a sample expansion and model construction method for spectroscopy-based chemometrics using limited soil samples.
• To achieve data augmentation and apply deep learning to original samples with inadequate numbers.

Main Methods:

• Developed a spectral subtraction method based on maximum or minimum values to acquire spectral and content differences.
• Implemented data augmentation by creating enhanced samples and expanding training or all data.
• Utilized Random Vector Functional Link (RVFL) network, Extreme Learning Machine (ELM), and 1D Convolutional Neural Network (1D CNN) for content prediction.

Main Results:

• The spectral subtraction method showed performance comparable to models using original samples.
• Augmenting all data improved performance for RVFL and ELM models.
• Augmenting training data demonstrated the superiority of 1D CNN over RVFL and ELM.

Conclusions:

• The proposed data augmentation method effectively expands samples for spectroscopy-based chemometrics with limited data.
• Deep learning models, especially 1D CNN, benefit significantly from this augmentation for soil analysis.
• This research provides a foundation for improved spectral preprocessing and modeling in precision agriculture.