For　received　signal　strength　(RSS)　fluctuations,　multipath　and　data　redundancy　issues　based　on　Wi-Fi　fingerprinting,　this　paper　proposes　an　indoor　localization　system　(SSMLoc),　which　combines　stacked　denoising　autoencoders　(SDAE)　and　multi-layer　perceptron　(MLP)　with　self-Attention　mechanism　(SAM).　The　system　uses　SDAE　to　generate　a　deep　robust　representation　of　RSS　data,　which　is　subsequently　used　to　build　a　localized　fingerprint　database.　A　simplified　SAM　reassigns　various　weights　to　the　internal　properties　of　deep　Wi-Fi　fingerprints,　allowing　the　MLP　network　to　better　capture　the　critical　features　of　Wi-Fi　fingerprints.　The　performance　of　the　SSMLoc　system　using　the　public　dataset　UJIIndoorLoc　is　validated　in　this　paper,　and　the　experimental　findings　reveal　that　the　system　has　great　robustness　and　enhances　positioning　accuracy　and　stability.　©　2022　IEEE.