Jiang et al. (2026) Non-destructive identification of closely related japonica rice seeds using hyperspectral imaging combined with HHO-optimized XGBoost

Identification

• Journal: Frontiers in Plant Science
• Year: 2026
• Date: 2026-09-17
• Authors: Chunxu Jiang, Shixue Gu, Gang Liu, Yang Liu, Rongjie Huang, Shaozhong Song
• DOI: 10.3389/fpls.2026.1909987

Research Groups

• School of Computer Science, Baicheng Normal University, China
• Haiming Primary School, Baicheng, China
• School of Data Science and Artificial Intelligence, Jilin Engineering Normal University, Changchun, China
• Rice Research Institute of Jilin Academy of Agricultural Sciences, China

Short Summary

This study presents a non-destructive identification method for closely related japonica rice seeds using hyperspectral imaging combined with HHO-optimized XGBoost. The approach achieves 94.72% overall accuracy and a Kappa coefficient of 0.9389.

Objective

• Develop a high-accuracy, non-destructive seed variety recognition technique for quality inspection in the seed industry.
• Identify closely related japonica rice seeds using hyperspectral imaging combined with machine learning algorithms.

Study Configuration

• Spatial Scale: Individual seed samples were analyzed using hyperspectral imaging technology.
• Temporal Scale: The study focused on a single time point, with no temporal analysis performed.

Methodology and Data

• Models used: XGBoost with Harris Hawks Optimization (HHO) for hyperparameter searching.
• Data sources: Hyperspectral images of 1800 japonica rice seed samples acquired using a Specim FX10 hyperspectral camera.

Main Results

• The HHO-XGBoost approach achieved an overall accuracy of 94.72% and a Kappa coefficient of 0.9389 for distinguishing nine closely related japonica rice varieties.
• SHAP analysis revealed the importance of critical spectral bands in the classification process.

Contributions

• This study provides a novel, non-destructive seed variety recognition technique using hyperspectral imaging combined with machine learning algorithms.
• The approach offers high accuracy and interpretability, addressing the limitations of traditional methods.

Funding

• No funding information is provided in the paper.

Citation

@article{Jiang2026Nondestructive,
  author = {Jiang, Chunxu and Gu, Shixue and Liu, Gang and Liu, Yang and Huang, Rongjie and Song, Shaozhong},
  title = {Non-destructive identification of closely related japonica rice seeds using hyperspectral imaging combined with HHO-optimized XGBoost},
  journal = {Frontiers in Plant Science},
  year = {2026},
  doi = {10.3389/fpls.2026.1909987},
  url = {https://doi.org/10.3389/fpls.2026.1909987}
}