Fault　detection　of　electrical　poles　is　part　of　the　daily　operation　of　power　utilities　to　ensure　the　sustainability　of　power　transmission.　This　paper　develops　a　method　for　intelligent　detection　of　fallen　poles　based　on　the　improved　YOLOX.　The　hyper-parameters　in　this　method　are　optimized　automatically　by　Particle　Swarm　Optimization　(PSO)　including　batch　size　and　input　resolution.　During　parameter　optimization,　a　specific　comprehensive　evaluation　metric　is　presented　as　the　fitness　function　to　obtain　optimal　solutions　with　low　labor　cost　and　high　method　performance.　In　addition,　virtual　pole　images　are　generated　by　3D　Studio　Max　to　overcome　the　imbalance　problem　of　normal　and　fault　data.　The　results　show　that　the　proposed　method　can　achieve　95.7%　of　recall　and　98.9%　of　precision,　which　demonstrates　the　high　accuracy　of　the　method　in　fallen　pole　detection.　In　the　comparative　experiment,　the　proposed　PSO-YOLOX　method　is　superior　to　the　existing　methods　including　original　YOLOX　and　Faster　R-CNN,　which　verifies　the　effectiveness　of　automatic　optimization　and　virtual　data　augmentation.　©　2024,　The　Author(s),　under　exclusive　licence　to　Springer　Science+Business　Media,　LLC,　part　of　Springer　Nature.