Hardware　is　the　foundation　of　the　Cyber-Physical　Systems,　while　the　security　threats　brought　by　the　hardware　Tro-jan　can　make　disastrous　consequences.　There　are　various　studies　on　detecting　hardware　Trojans,　but　in　order　to　defend　against　the　hardware　Trojans,　the　researches　just　staying　at　the　shallow　detection　stage　are　far　away　from　the　goal.　Finding　the　specific　location　of　hardware　Trojans　is　a　prerequisite　for　more　precise　control　of　them.　In　this　paper,　a　hardware　Trojans　localization　method　utilizing　deep　learning　at　the　gate-level　(RLocHT)　is　proposed.　RLocHT　is　based　on　the　hardware　Trojans　detection　results　of　TextCNN,　combined　with　the　path　division　technology,　which　lets　the　convolutional　neural　network　detect　the　divided　paths　in　more　detail　to　narrow　the　range　of　locating　hardware　Trojans　and　achieve　the　purpose　of　localization.　In　addition,　two　new　evaluation　metrics,　localization　loss　rate　(LoLR),　and　total　locating　deviation　(SD_{Delta}),　are　defined　to　evaluate　the　integrity　and　accuracy　of　RLocHT.　Finally,　the　length　of　the　division　cutlen　in　the　path　division　technology　is　also　adjusted　to　optimize　the　performance.　With　the　optimal　parameter　of　each　netlist,　for　all　seven　benchmarks,　the　results　achieve　an　average　localization　loss　rate　LoLR_{avg}　of　33.28%　and　an　average　total　locating　deviation　SD_{Delta　avg}　of　2.55.　©　2022　IEEE.