Lithium-ion batteries are widely used in electric vehicles and energy storage systems due to their high energy density and long cycle life. However, repeated charge–discharge cycles inevitably lead to battery degradation, resulting in a reduction of the remaining useful life (RUL). Therefore, accurate diagnosis of the state of health (SoH) is essential for reliable operation and efficient maintenance of batteries. The accuracy of SoH diagnosis strongly depends on the selection of informative and reliable health indicators (HI). Peaks observed in incremental capacity curves are closely associated with lithium inventory loss and loss of active material, and thus effectively reflect battery degradation behavior. However, Peaks of incremental capacity curves rely on heuristic feature selection, which limits their ability to autonomously identify voltage regions that are most relevant to degradation. In contrast, attention mechanisms enable the learning of feature importance within peaks, allowing artificial intelligence models to automatically select degradation-sensitive health indicators. In this work, the extracted HI are used to diagnose SoH and predict RUL of lithium-ion batteries. Incremental capacity curves are computed from publicly available charge discharge datasets, and correlation analysis is performed to select the most informative indicators. These selected HI are then employed as input features for a Gaussian Process Regression model for SoH diagnosis and RUL prediction. The robustness of the proposed framework is validated under different cycle number conditions and across multiple battery datasets, demonstrating its effectiveness and generalization capability.