Deep　learning-based　Wi-Fi　fingerprint　localization,　which　needs　to　receive　signal　strength　(RSS)　fingerprint　for　training,　has　been　widely　studied.　Some　inherent　problems　impact　localization　accuracy,　such　as　device　heterogeneity　and　effective　feature　extraction.　Effective　feature　extraction　and　strict　location　estimation　mechanisms　can　significantly　improve　the　accuracy　of　localization.　In　this　paper,　we　propose　an　indoor　localization　method　based　on　multilevel　feature　extraction　and　autoregressive　location　estimation　to　further　capture　the　intrinsic　features　of　RSS　(Received　Signal　Strength),　and　learn　the　approach　of　mapping　fingerprint　features　to　location　information.　Experimental　results　show　that　the　proposed　method　achieves　higher　localization　accuracy　compared　with　DNN　and　CNN-based　fingerprint　localization　methods.　In　the　multi-building　scenario,　the　proposed　method　achieves　higher　localization　accuracy　than　the　compared　baseline　methods.　©　2023　IEEE.