Since　integrated　circuits　are　performed　by　several　untrusted　manufacturers,　malicious　circuits　(hardware　Trojans)　can　be　implanted　in　any　stage　of　the　Internet-of-Things　(IoT)　devices.　With　the　globalization　of　the　IoT　device　manufacturing　technologies,　protecting　the　system-on-chip　(SoC)　security　is　always　the　keys　issue　for　scientists　and　IC　manufacturers.　The　existing　SoC　high-level　synthesis　approaches　cannot　guarantee　both　register-transfer-level　and　gate-level　security,　such　as　some　formal　verification　and　circuit　characteristic　analysis　technologies.　Based　on　the　structural　characteristics　of　hardware　Trojans,　we　propose　a　multi-layer　hardware　Trojan　protection　framework　for　the　Internet-of-Things　perception　layer　called　RG-Secure,　which　combines　the　third-party　intellectual　property　trusted　design　strategy　with　the　scan-chain　netlist　feature　analysis　technology.　Especially　at　the　gate　level　of　chip　design,　our　RG-Secure　is　equipped　with　a　distributed,　lightweight　gradient　lifting　algorithm　called　lightGBM.　The　algorithm　can　quickly　process　high-dimensional　circuit　feature　information　and　effectively　improve　the　detection　efficiency　of　hardware　Trojans.　In　the　meanwhile,　a　common　evaluation　index　F-measure　is　used　to　prove　the　effectiveness　of　our　method.　The　experiments　show　that　RG-Secure　framework　can　simultaneously　detect　register-transfer-level　and　gate-level　hardware　Trojans.　For　the　trust-HUB　benchmarks,　the　optimized　lightGBM　classifier　achieves　up　to　100%　true　positive　rate　and　94%　true　negative　rate;　furthermore,　it　achieves　99.8%　average　F-measure　and　99%　accuracy,　which　shows　a　promising　approach　to　ensure　security　during　the　design　stage.