Electrooculogram　(EOG)　is　an　inevitable　main　interference　in　electroencephalogram　(EEG)　acquisition,　which　directly　affects　the　analysis　and　application　of　EEG.　Second-order　blind　identification　(SOBI),　as　a　blind　source　separation　(BSS),　has　been　used　to　remove　the　ocular　artifacts　(OA)　of　contaminated　EEG.　However,　SOBI　that　assumes　the　source　signal　to　be　stationary　is　not　appropriate　for　nonstationary　EEG　signals,　yielding　undesirable　separation　results.　In　addition,　it　is　regrettable　that　the　current　discriminations　of　ocular　artifacts,　such　as　correlation　coefficients,　sample　entropy,　do　not　take　into　account　of　the　fuzzy　characteristics　of　EEG,　which　leads　to　the　inaccurate　judgement　of　OA.　In　this　paper,　a　novel　OA　removal　method　is　proposed　based　on　the　combination　of　discrete　wavelet　transform　(DWT)　and　SOBI　and　denoted　as　DWSOBI.　DWT　is　used　to　analyze　each　channel　of　contaminated　EEG　to　obtain　more　stable　multi-scale　wavelet　coefficients;　then,　the　wavelet　coefficients　in　the　same　layer　are　selected　to　construct　the　wavelet　coefficient　matrix,　and　it　is　further　separated　by　using　SOBI　to　obtain　the　estimation　of　source　signals,　whose　fuzzy　entropies　are　calculated　and　employed　to　realize　the　automatic　identification　and　removal　of　OA.　Based　on　a　public　database,　many　experiments　are　conducted　and　two　performance　indexes　are　adopted　to　measure　the　elimination　effect　of　OA.　The　experiment　results　show　that　DWSOBI　achieves　more　adaptive　and　accurate　performance　for　four　kinds　of　OA　from　three　subjects,　and　is　superior　to　the　commonly　used　methods.　©　2021,　The　Editor(s)　(if　applicable)　and　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.