Electrode microstructure-driven multi-scale feature fusion framework based on microscopic imaging technique for health estimate of lithium-ion batteries

Accurately estimating the health of lithium-ion batteries is crucial for ensuring their long-term reliability and stable performance. Traditional estimating methods, primarily based on electrical and thermal signal analyses, do not account for microstructural changes predominantly characteristic of capacity loss and performance degradation. Therefore, this study introduces a novel health estimation framework based on scanning electron microscopy integrated with cross-modal fusion to extract morphological and high-dimensional structural features. Pyramid-attention and region-adaptive fusion strategies are developed to combine information obtained at different magnifications and from different electrode regions, thereby balancing detailed microstructural characteristics and overarching degradation trends. Experimental results indicate that this hybrid feature fusion approach, incorporating data augmentation and synthetic sample generation, improves estimation accuracy and model generalization. The pyramid-attention mechanism optimizes multi-magnification information integration, and region-adaptive weighting enhances the degradation assessment stability. The proposed model outperforms single-modality approaches and achieves a root mean squared error of 1.25% under optimal conditions, with the enhancements contributing to an error reduction of up to 65.7% compared with that under non-enhanced datasets. The proposed method leverages microstructural insights to offer a scalable solution for battery health monitoring and material development and a deeper understanding of degradation mechanisms and battery performance improvements.