Subject to variable operating conditions, sensor noise and capacity regeneration in real‑world operation of batteries, the accurate extraction of degradation features from complex discharge signals remains a core challenge for battery state‑of‑health (SOH) estimation. In this work, a lightweight Convolutional Neural Network‑Vision Transformer (CNN-ViT) hybrid network, denoted as GAF‑MEF‑ViT, is proposed for lithium‑ion battery SOH estimation, which implements multi‑encoded feature fusion based on the Gramian Angular Field. First, the multi‑channel one‑dimensional battery signals are individually normalized and then converted into Gramian Angular Summation Field (GASF) and Gramian Angular Difference Field (GADF) images. Second, a lightweight CNN‑ViT hybrid encoder is designed. The CNN branch is responsible for extracting local degradation texture features, while the ViT branch models the global dependencies within images. Further, a multi‑encoded feature gating fusion module is constructed to adaptively learn the importance weights of GASF and GADF. A regression module is then utilized to accomplish lithium‑ion battery SOH estimation. Finally, experimental validations are carried out on the NASA lithium‑ion battery dataset. The experimental results demonstrate that the proposed method can effectively track the battery capacity degradation trend and achieve favorable SOH estimation performance.

Lithium-ion battery,state of health,image coding,feature fusion,hybrid network