Identification of maize kernel varieties based on interpretable ensemble algorithms.

Area of Science:

• Agricultural Science
• Computer Science
• Data Science

Background:

• Maize kernel variety identification is vital for reducing storage losses and ensuring food security.
• Traditional single models struggle with large-scale multimodal data for maize identification.

Purpose of the Study:

• To construct an interpretable ensemble learning model for maize seed variety identification.
• To address limitations of traditional models in processing multimodal data.

Main Methods:

• Utilized multimodal data fusion (morphological and hyperspectral data).
• Developed an improved differential evolutionary algorithm for parameter optimization.
• Employed a Stacking integration model with optimized base learners.
• Applied Shapley Additive exPlanation for model interpretability.

Main Results:

• The HDE-Stacking identification model achieved 97.78% accuracy.
• Identified key spectral bands (784 nm, 910 nm, 732 nm, 962 nm, 666 nm) influencing identification.

Conclusions:

• The developed model offers a scientific basis for efficient and accurate corn kernel variety identification.
• Enhances traceability in germplasm resource management and improves agricultural quality management for food security.