At　present,　X-ray　imaging　technology　is　widely　used　in　the　fields　of　security　inspection　and　medical　treatment,　and　plays　a　vital　role　in　the　safety　of　people＇s　lives　and　property.　This　paper　mainly　does　in-depth　research　on　the　detection　of　suspicious　objects　in　the　field　of　security　inspection　by　X-ray,　which　has　made　a　great　contribution　to　improving　the　efficiency　and　accuracy　of　security　inspection.　In　recent　years,　the　convolutional　neural　network　has　made　great　breakthroughs　in　the　field　of　computer　vision,　making　it　possible　to　detect　objects　in　real-time　X-ray　images.　However,　due　to　the　overlap　and　occlusion　of　objects　in　X-ray　images　compared　with　natural　images,　the　convolutional　neural　network　does　not　achieve　good　training　results.　In　this　paper,　we　propose　an　object　detection　algorithm　based　on　the　SSD　model,　using　the　improved　HarDNet　network　as　the　backbone　network,　and　introducing　a　multi-scale　feature　fusion　mechanism　and　an　attention　mechanism　to　improve　the　detection　accuracy.　The　multi-scale　feature　fusion　module　combines　the　feature　maps　of　different　levels　in　the　backbone　network,　and　fuses　the　low-level　texture　information　and　high-level　semantic　information　to　obtain　more　effective　feature　maps.　In　addition,　the　attention　module　constructs　two　sub-modules:　the　spatial　attention　module　and　the　channel　attention　module,　respectively　summarize　the　attention　information　of　the　space　and　the　channel,　and　synthesize　the　information　to　obtain　more　comprehensive　and　reliable　attention　information.　We　evaluate　our　method　extensively　on　HiXray　and　SIXray　and　the　results　demonstrate　that　it　outperforms　SOTA　detection　methods.　©　2023　IEEE.