The　Tooth　mark　is　an　important　attribute　of　tongue　diagnosis　in　Traditional　Chinese　Medicine　(TCM).　The　recognition　results　obtained　by　current　methods　heavily　rely　on　the　results　of　tongue　image　segmentation.　To　solve　the　problem,　we　regarded　the　tooth　marks　recognition　as　a　task　of　object　detection　and　improved　the　original　single　shot　detector　(SSD)　to　detect　the　tooth　marks.　We　removed　the　last　two　prediction　layers　of　SSD　and　set　the　aspect　ratios　of　the　prior　box　to　1　based　on　the　statistical　data　of　the　size　and　aspect　ratios　of　tooth　mark　regions.　Then　we　designed　the　multiple　feature　fusion　module　to　combine　the　multi-scale　features　and　embedded　them　hierarchically　into　the　network　to　transfer　the　semantic　information　from　deep　layers　to　shallow　layers.　Furthermore,　we　also　developed　a　feature　enhancement　module　to　improve　the　distinctiveness　of　features.　The　experimental　results　showed　that　the　proposed　method　achieved　96.8%　in　terms　of　accuracy,　which　is　significantly　better　than　the　current　methods.　©　2022,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.