<p>Accurate evaluation of the state of health (SOH) of lithium-ion batteries (LIBs) is crucial for ensuring the stability and reliability of energy storage and power systems. In practical applications, however, SOH estimation often encounters challenges such as missing data, noise interference, and limited labeled samples. To address these issues, a multi-channel Gramian Angular Field (GAF) image modeling and self-supervised Forgetting Vision Transformer framework (GAF-SSFViT) based on short-term relaxation voltage is proposed. The proposed approach integrates GAF-based image encoding, Simple Siamese Representation (SimSiam) self-supervised contrastive learning, Vision Transformer (ViT) feature extraction, and Long Short-Term Memory (LSTM) temporal modeling to achieve high-precision SOH prediction and dynamic characterization. Through image augmentation strategies such as random masking and channel dropout, the model effectively captures degradation-sensitive representations during unsupervised pretraining and achieves end-to-end fine-tuning via progressive layer unfreezing. Comparative experiments were conducted and validated on both a custom dataset and the public dataset released by Xi’an Jiaotong University (XJTU), demonstrating the superiority of the proposed method. The results indicate that the new approach achieves significantly higher estimation accuracy and robustness than other models, reducing the Mean Squared Error (MSE) by approximately 60% on average, thereby fully verifying the advantages of the GAF-SSFViT framework. This study provides a robust and efficient SOH estimation framework for intelligent battery management systems (BMS).</p>