The　radiographic　test　used　to　inspect　the　internal　defects　is　an　important　non-destructive　testing　technique　for　welding.　The　weld　bead　radiographs　of　different　detection　scenes　and　different　types　of　defects　are　very　different,　which　limits　the　versatility　of　the　automatic　detection　algorithm.　This　paper　solves　this　problem　by　analyzing　the　radiographic　images　at　different　scales　and　intensities.　Firstly,　a　multi-scale　multi-intensity　parameter　space　is　established,　and　the　preprocessed　images　corresponding　to　the　parameters　ensure　that　defects　are　not　missed.　Then,　through　the　detection　standard　of　the　weld　and　the　properties　of　the　radiographic　image,　the　value　range　of　the　parameters　is　automatically　limited　and　the　preprocessed　image　is　optimized.　Finally,　algorithms　for　screening　and　merging　defects　in　different　preprocessed　images　are　designed　to　reduce　false　detections　and　describe　precise　defect　boundaries.　Experimental　results　show　that　the　algorithm　proposed　in　this　paper　is　universal,　robust　and　accurate.