The estimation of state of health (SOH) for lithium-ion batteries is considered a critical task to ensure battery reliability and enable accurate lifetime prediction. To address the limitations of existing SOH estimation methods in handling outliers and selecting effective features, a novel health state feature processing approach is proposed. An improved Hampel filter with adaptive thresholding and multidimensional feature fusion optimization is integrated. The method is composed of three stages. In the initial feature selection stage, ten candidate features are extracted from three perspectives: the charging process, the discharging process, and the capacity increment curve. In the feature correction stage, an adaptive-threshold Hampel filtering algorithm is developed, which dynamically adjusts the window size and threshold to correct abnormal feature values. In the feature selection stage, dual-correlation analysis using both Pearson and Spearman coefficients is employed to identify key features. Finally, using the NASA battery dataset, a gated recurrent unit (GRU) network is adopted to evaluate the performance of different feature combinations for SOH estimation. Simulation results demonstrate that the selected three-feature combination significantly improves the accuracy and stability of SOH estimation.