Accurate estimation of the lithium-ion batteries (LiBs) state of health (SOH) is vital for ensuring the safety and dependability of electric vehicles (EVs) throughout their life cycle. However, the traditional neural network estimation has low fitting accuracy, and most research in this field is based solely on normal temperature states. To address this, we propose extracting health factors (HFs) from battery data under different temperature states, employing the Gaussian process regression algorithm (GPR) to estimate SOH. We also examine the influence of various kernel functions on the estimation results of the GPR algorithm.