Identification of maize kernel varieties based on interpretable ensemble algorithms

• DOI: 10.3389/fpls.2025.1511097
• Publication Date: 2025-02-12T07:30:38Z
• Authors: Chunguang Bi; Xinhua Bi; Jinjing Liu; Hao Xie; Shuo Zhang; He Chen; Mohan Wang; Lei Shi; Shaozhong Song
• Keywords: Identification; Kernel (algebra); Germ plasm; Ensemble Learning; Traceability

Introduction Maize kernel variety identification is crucial for reducing storage losses and ensuring food security. Traditional single models show limitations in processing large-scale multimodal data. Methods This study constructed an interpretable ensemble learning model maize seed through improved differential evolutionary algorithm data fusion. Morphological hyperspectral of samples were extracted preprocessed, three methods used to screen features, respectively. The base learner the Stacking integration was selected using diversity performance indices, with parameters optimized a evolution incorporating multiple mutation strategies dynamic adjustment factors recombination rates. Shapley Additive exPlanation applied learning. Results HDE-Stacking achieved 97.78% accuracy. spectral bands at 784 nm, 910 732 962 666 nm showed positive impacts on results. Discussion research provides scientific basis efficient different corn varieties, enhancing accuracy traceability germplasm resource management. findings have significant practical value agricultural production, improving quality management efficiency contributing security assurance.