Accurate state of health (SOH) estimation of lithium-ion batteries (LIBs) is critical to ensuring the safety and reliability of battery management system (BMS). To achieve precise estimation, this study proposes a hybrid framework that integrates variational mode decomposition (VMD), kernel principal component analysis (KPCA), polar lights optimizer (PLO), and physics-informed neural network (PINN) for SOH estimation. First, multidimensional health features are extracted and decomposed by VMD into intrinsic mode functions (IMFs), which are then compressed into a one-dimensional principal component via KPCA, retaining over 95% of the original information. Subsequently, the PLO algorithm is used to adaptively optimize three key hyperparameters of the PINN-based model: the learning rate, the number of collocation points, and the regularization loss weight. Finally, the optimized PINN is deployed to predict the SOH of the Center for Advanced Life Cycle Engineering (CALCE) battery dataset. Experimental results demonstrate that the proposed VMD-KPCA-PLO-PINN exhibits high prediction accuracy under both 7:3 and 5:5 training-to-testing data partitions. For example, under the 5:5 partition, the proposed model achieves an average $R^2$ of 0.983 and an average RMSE of 0.0085 on the tested CALCE cells.