The invention discloses a detection method of a lithium battery SOH (status of health). The detection method comprises the following steps: 1) establishing an equivalent circuit model of a lithium battery, wherein the equivalent circuit model comprises a branch composed of an open-circuit voltage Uoc, a polarized capacitor Cp, an Ohmic internal resistor Ro in serial connection in order; two ends of the open-circuit Uoc are connected with self-discharging resistors Rd in parallel, two ends of the polarized capacitator are connected with polarized internal resistors Rp in parallel, thereby obtaining an equation of the equivalent circuit model of the lithium battery; 2) establishing a discrete system state space equation and observation equation of the battery SOC, the Ohmic internal resistorRo and the self-discharging internal resistor Rd according to the equivalent circuit model equation of the lithium battery; performing iteration computation by adopting an extended Kalman filter algorithm to obtain estimation values of the battery SOC, the Ohmic internal resistor Ro and the self-discharging internal resistor Rd; and 3) regarding the estimation values of the battery SOC, the Ohmicinternal resistor Ro and the self-discharging internal resistor Rd and the cycle index H of the lithium battery as input variables to input into a neural network model, and estimating to obtain the lithium battery SOH.