Facial　expression　recognition　(FER)　is　a　challenging　task　due　to　various　unrestricted　conditions.　Normal　facial　expression　algorithms　work　well　on　frontal　faces.　However,　detection　expression　from　the　occluded　faces　is　still　a　challenging　task.　In　this　paper,　we　propose　a　novel　deep　convolution　neural　network　with　self-attention　mechanism　in　order　to　detect　the　occlusion　regions　in　the　face　for　efficient　recognition　of　facial　expression.　Firstly,　we　use　a　backbone　CNN　to　extract　feature　maps　of　the　facial　images.　Then　the　global　self-attention　with　relative　position　encodings　is　utilized　to　process　and　aggregate　the　information　contained　in　the　feature　maps.　The　global　self-attention　can　learn　the　relationships　between　the　single　feature　and　entire　facial　information.　Inorder　to　pay　attention　to　the　highly　relative　region　and　ignore　the　information-deficient　regions.　To　evaluate　the　efficiency　of　the　proposed　model,　we　use　two　wild　benchmark　datasets　RAF　and　AffectNet　for　FER.　The　results　show　that　proposed　model　is　effective　to　recognize　the　facial　expression　from　the　occluded　images.　©　The　Author(s),　under　exclusive　license　to　Springer　Nature　Switzerland　AG　2023.