The state of health(SOH) of lithium-ion battery directly determines the ability to store energy and output power. When the transportation equipped with lithium-ion battery is running, the battery data needs to be uploaded in real time. The higher the data recording frequency, the higher the data communication cost. In order to ensure the accuracy of battery SOH estimation and reduce the data communication cost, a wide sampling frequency charge-discharge experiment with different charge-discharge rates was designed. For the fluctuation of health features of wide sampling frequency, locally weighted linear regression(LWLR) algorithm was used to qualitatively characterize the health features decline trend. Maximum information coefficient(MIC) algorithm was used to measure the correlation between health features and capacity. Finally Bi-directional long and short-term memory (BI-LSTM) based neural network further learned the nonlinear degradation relationship between capacity and health features. Estimating the battery SOH offline was conducted based on the single battery historical data, and the maximum relative error was 1.601%.