Abstract: Accurate state of health (SOH) estimation is crucial for battery management systems, yet challenging due to the nonlinear and multi-scale coupling characteristics of aging. To address this issue, CTANet, a novel model integrating CNN, TimesNet, and a dual-domain strip attention mechanism (DSAM) to capture complex intra- and inter-series dependencies, is proposed. First, health indicators (HIs) are extracted and optimized using the 3σ rule and Pearson correlation analysis. Subsequently, the model employs CNN for feature extraction, utilizes the TA-Block to capture multi-scale temporal dependencies, and leverages DSAM to adaptively weight critical information across spatial and frequency domains. Validation experiments on MIT and XJTU datasets demonstrate that the proposed method achieves stable, high-precision estimation. The results show errors consistently within 2% and mean absolute errors (MAE) below 0.33%, providing robust support for effective battery health management.