A joint state of charge (SOC)-state of health (SOH) method of estimation was proposed in order to improve the state estimation accuracy of unmanned underwater vehicle (UUV) battery management system. A test bench was constructed, and four groups of lithium-ion batteries were used for charging and discharging test under the whole life cycle. Data under different attenuation degrees were obtained. Four-dimensional factors were designed by theoretical derivation and experimental analysis, and a SOH estimation model based on improved support vector regression (SVR) was established. The coupling relationship between battery states was explored. A SOC estimation model based on extended Kalman filter (EKF) was established and the forgetting factor recursive least squares (RLS) algorithm was used to update the model parameters. The SOC estimation results were corrected by SOH. The method was validated through different testing conditions experiment. Results show that the four-dimensional characterization factor and battery capacity have good correlation. The accuracy of SOH estimation model is high, and the accuracy of SOC estimation model is improved by joint modification. The proposed joint estimation method has high universality and reliability, and can be used as an effective state estimation algorithm for embedded battery management system.