The　rise　of　unmanned　supermarkets　highlights　a　growing　trend　towards　complete　automation　in　shopping　activities　in　the　future.　In　unmanned　supermarkets,　artificial　intelligence　plays　a　pivotal　role　in　fruit　vision　recognition　for　pricing,　which　reduces　labor　costs　compared　to　traditional　supermarkets,　where　manual　identification　and　pricing　of　fruits　by　cashiers　are　commonplace.　However,　existing　intelligent　fruit　recognition　models　often　neglect　the　influence　of　packaging　on　precise　identification.　To　meet　the　demand　for　accurate　recognition　of　packaged　fruits　in　retail　settings,　this　study　investigates　deep　learning　techniques　for　the　detection　and　classification　of　fruits　packaged　in　commonly　used　transparent　plastic　bags　or　boxes　in　supermarkets　and　grocery　stores.　Four　intelligent　recognition　and　classification　models,　namely　Support　Vector　Mashine(SVM),　Visual　Geometry　Group　Network　16(VGG16),　MobileNetV3-Small　and　You　Only　Look　Once　X-Middle(YOLOX-M)　are　trained　and　compared　using　a　dataset　created　for　this　research,　which　contains　nearly　10,000　images　of　eight　widely　consumed　fruits　with　packages.　The　dataset　is　meticulously　labeled　and　augmented,　and　the　performance　of　each　model　is　evaluated　using　five　metrics,　precision　(P),　recall　(R),F1-score　(F1),　mean　average　precision　(mAP),　and　average　detection　time　(t)　per　image.　The　comprehensive　evaluation　results　reveal　YOLOX-M　as　the　superior　model,　displaying　an　average　recognition　accuracy　of　98.74%　and　an　average　detection　time　of　6ms　per　image.　These　findings　underscore　the　high　potential　of　YOLOX-M　for　application　in　retail　settings.　Our　research　contributes　to　the　development　of　intelligent　fruit　pricing　devices　and　promotes　fruit　species　recognition　with　intelligent　systems　for　unmanned　supermarkets.　©　2024　IEEE.