The　flexibility　of　small　UAVs　allows　them　to　be　deployed　in　a　variety　of　emerging　markets,　while　the　intrusion　of　uncertified　UAVs　into　sensitive　areas　poses　a　great　threat.　In　recent　years,　radio　frequency　fingerprint　identification　technology,　as　a　new　physical　layer　authentication　technology,　has　been　widely　used　in　UAV　identification.　It　is　difficult　for　traditional　radio　frequency　identification　technology　to　meet　the　needs　of　massive　data.　Deep　neural　networks　generally　have　deep　layers　and　require　a　large　amount　of　computing,　which　makes　it　difficult　to　be　deployed　in　edge　devices　with　limited　computing　power.　In　this　paper,　a　lightweight　neural　network　for　radio　frequency　fingerprint　identification　of　UAVs　is　designed.　In　the　backbone　network,　multi-scale　Res2Net　modules　are　used　to　replace　the　standard　bottleneck.　To　further　improve　model　performance,　an　Efficient　Channel　Attention　(ECA)　block　is　embedded　in　the　Res2Net　module.　As　for　data　processing,　to　promote　the　generalization　of　the　model,　Mixup　is　used　to　enhance　the　data.　The　results　show　that　the　proposed　lightweight　method　can　effectively　identify　unmanned　aerial　vehicles,　with　the　average　accuracy　rate　reaching　93%　within　SNR　[−5,　10]　dB.　Meanwhile,　the　performance　is　better　than　other　methods　with　low　signal-to-noise　ratio.　©　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.　2025.