Because　of　its　intra-class　variance,　food　image　recognition　is　a　very　challenging　task.　In　this　paper,　we　present　an　automated　food　classification　system　that　can　recognize　24　different　types　of　Senegalese　food　from　pictures.　First,　a　new　small-scale　food　picture　dataset　of　the　most　common　local　Senegalese　foods　named　FoodNet221　was　built,　which　contains　6028　food　photos　with　24　categories.　Second,　a　method　suitable　for　food　image　classification　on　small　data　is　proposed.　This　method　use　an　updated　EfficientNet　model　modified　by　extending　the　EfficientnetB0　from　Mingxing　Tan　and　Quoc　V.Le　to　the　task　of　food　recognition　through　transfer　learning.　Experiments　show　that　the　method　is　particularly　well　suited　to　Senegalese　food　dataset　FoodNet221,　with　training　accuracy　of　97.95%.　©　2021　ACM.