Data-driven　methods　have　been　widely　used　to　estimate　the　State　of　health　(SOH)　of　Lithium-Ion　batteries　(LIBs).　However,　these　methods　lack　interpretability.　In　response　to　this　issue,　this　article　proposes　a　method　called　Physics-informed　neural　network　(PIFNN)　to　enhance　the　interpretability　of　predictions　made　by　a　feedforward　neural　network　(FNN).　First,　the　features　are　extracted　from　incremental　capacity　(IC)　curves　and　differential　temperature　curves,　which　can　characterize　battery　aging　from　different　perspectives.　Specifically,　the　peaks　of　the　IC　curves　(P-IC)　reflect　the　electrochemical　reactions　that　occur　during　the　charge-discharge　processes　of　LIBs.　The　decline　of　the　P-IC　is　related　to　the　loss　of　active　materials　in　LIBs,　which　is　a　major　cause　of　the　decrease　of　the　SOH.　This　article　converts　the　monotonous　relationship　between　the　P-IC　and　the　SOH　into　physical　constraints　to　guide　the　＂learning　process＂　of　the　model.　In　the　prediction　process,　a　physics-constrained　secondary　＂training＂　is　applied　to　the　FNN　predictions　to　further　enhance　interpretability　and　improve　prediction　accuracy.　The　feasibility　of　the　proposed　method　is　validated　using　the　Oxford　and　NASA　battery　datasets.　The　results　indicate　that　PIFNN　effectively　improves　prediction　accuracy　and　reduces　errors　to　below　1.5　%.