In intelligent lithium-ion battery management, the state of health (SOH) of battery is essential for the batteries’ running in electric vehicles. Popularly, the battery SOH is estimated by using suitable features and data-driven methods. However, it is difficult to extract appropriate features characterizing battery SOH from the charging and discharging data of batteries owing to various state of charges (SOCs) and working conditions of batteries. In order to effectively estimate the battery SOH, an estimation method based on gradual decreasing current, double correlation analysis and gated recurrent unit (GRU) is proposed in this paper. Firstly, gradual decreasing current in the constant voltage charging phase is measured as the raw data. Then, the double correlation analysis method is proposed to select combined features characterizing the battery SOH from different categories of features. Meanwhile, the number of input features is also ensured by the method. Finally, the GRU algorithm is employed to set up a SOH estimation model whose learning rate is improved by using a sparrow search algorithm (SSA) for the purpose of capturing the hidden relationship between features and SOH. The adaptability of the proposed method is validated by SOH estimation experiments of a single battery and a battery pack. Additionally, contrast experiments are performed to show the advanced estimation performance of the proposed method.