In　this　paper,　we　propose　a　deep　learning　framework　to　automatically　detect　pollen　grains　instead　of　the　manual　counting　of　pollen　numbers　under　an　optical　microscope.　Specifically,　we　first　establish　a　large-scale　dataset　of　pollen　grains,　which　contains　3000　images　of　five　subcategories.　All　the　images　in　our　dataset　are　scanned　by　an　optical　microscope.　Then,　a　pollen　grain　detector　(PGD)　based　on　deep　learning　is　designed　to　eliminate　the　effects　of　noise　and　capture　subtle　features　of　pollen　grains.　Finally,　extensive　experiments　are　conducted　and　show　that　the　proposed　PGD　method　achieves　the　best　performance　(84.52%　mAP).　©　2022,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.