The　internal　defects　of　complex　components　are　usually　detected　by　X-ray,　and　the　detection　images　generally　have　the　problems　of　large　gray　change　and　low　contrast.　The　research　on　image　enhancement　method　is　of　great　significance　to　improve　the　accuracy　of　defect　recognition.　To　overcome　the　problems　of　adjusting　parameters　and　the　influence　of　personnel　experience　in　traditional　enhancement　methods,　this　paper　proposes　a　radiographic　image　enhancement　method　based　on　deep　learning　theory.　Firstly,　according　to　the　requirements　of　radiographic　image　enhancement,　the　target　data　set　of　radiographic　image　enhancement　is　constructed.　Secondly,　using　the　deep　learning　theory,　an　improved　U-Net　network　image　enhancement　model　is　designed　to　realize　image　structure　preservation　and　noise　removal.　Finally,　the　proposed　method　is　illustrated　and　verified　by　the　radiographic　images　of　complex　metal　components.　©　2022　IEEE.