Accuracy　classification　of　flower　species　is　a　key　step　for　the　management　of　flower　breeding.　At　present,　convolutional　neural　networks　are　efficient　feature　representation　learning　algorithms　that　are　widely　used　to　improve　the　accuracy　of　flower　image　classification.　However,　there　are　some　limitations　in　selecting　appropriate　network　structures,　parameters,　and　algorithms,　and　in　the　requirement　for　huge　training　times　for　optimal　recognition　performance　in　real-world　applications.　To　overcome　these　difficulties,　we　developed　a　convolutional　neural　network　ensemble　method　for　flower　species　classification.　The　method　has　three　steps:　(1)　MobileNet　models　pre-trained　on　the　ILSVRC-2012-CLS　image　classification　dataset　were　used　as　the　single　classifiers　for　feature　extraction;　(2)　the　models　were　transferred　to　flower　datasets　to　train　several　different　classifiers　and　a　re-sampling　strategy　was　used　to　enhance　the　diversity　of　individual　models;　and　(3)　an　ensemble　model　was　constructed　using　the　weighted　average　method.　To　verify　the　effectiveness　of　the　proposed　method,　classification　experiments　were　performed　on　two　flower　image　datasets.　The　results　showed　that　the　proposed　convolutional　neural　network　ensemble　was　feasible　and　effective,　and　had　better　generalization　ability　and　higher　recognition　rate　than　the　single　classifiers.　The　convolutional　neural　network　ensemble　can　also　be　used　and　modified　for　the　classification　of　other　crop　images.　©　2020　ACM.