The　rapid　and　accurate　identification　of　tomato　diseases　is　the　basis　of　crop　disease　control.　In　order　to　achieve　accurate　identification　of　tomato　diseases,　this　paper　first　explores　the　impact　of　model　depth　and　the　presence　or　absence　of　mixup　data　enhancement　on　the　ResNet　model.　Experimental　results　show　that　using　the　data　set　enhanced　by　mixup　can　effectively　make　the　model　more　robust.　At　the　same　time,　ResNet34　depth　model　recognition　accuracy　is　higher.　Considering　the　differences　in　classification　accuracy　and　calculation　speed　between　ResNet　and　SE-ResNet,　the　paper　chooses　the　SE-ResNet　model　as　the　basis　of　the　network　structure.　We　tuned　the　model　and　built　a　SE-ResNet　network　that　is　more　suitable　for　tomato　disease　identification.　The　experimental　results　show　that　the　accuracy　of　training　SE-ResNet　using　datasets　such　as　mixup　is　88.83%.　Our　model　can　effectively　identify　various　tomato　diseases　and　the　severity　of　each　tomato　disease.　©　2020　IOP　Publishing　Ltd.　All　rights　reserved.