In　order　to　accurately　transmit　the　content　meaning　of　vibrotactile　signals　and　achieve　intelligent　recognition　and　signal　reconstruction,　a　joint　vibrotactile　coding　scheme　for　machine　recognition　and　human　perception　was　proposed.　At　the　encoding　end,　the　original　three-dimensional　vibrotactile　signals　were　converted　into　one-dimensional　signals.　Then　the　semantic　information　of　the　signals　was　extracted　using　a　short-time　Fourier　transform　before　being　effectively　compressed　and　transmitted.　At　the　decoding　end,　a　fully　convolutional　neural　network　was　used　to　intelligently　recognize　based　on　the　semantic　information.　The　difference　between　the　original　signals　and　the　reconstructed　signals　based　on　semantic　information　was　used　as　compensation　for　the　semantic　information,　and　the　quality　of　the　reconstructed　signals　was　gradually　improved　to　meet　human　perceptual　needs.　The　experimental　results　show　that　the　proposed　scheme　achieve　tactile　recognition　with　semantic　information　at　a　lower　bit　rate　while　improving　the　compression　efficiency　of　tactile　data,　thus　satisfying　human　perceptual　needs.　©　2023　Editorial　Board　of　Journal　on　Communications.　All　rights　reserved.