With　the　development　of　robot　technology,　intelligent　robot　has　been　widely　used　in　substation　inspection.　However,　the　current　target　detection　algorithm　faces　the　problem　of　too　many　parameters　to　realize　real-time　detection　on　embedded　platform.　To　address　this　issue,　an　improved　YOLOv5　algorithm　based　on　AI　front-end　is　proposed　for　substation　instrument　detection.　The　algorithm　is　based　on　YOLOv5　network　and　introduces　the　SE　fusion　attention　mechanism　module,　which　adaptively　learns　the　relationship　between　feature　channels　to　improve　the　model＇s　ability　to　extract　important　features　of　objects.　At　the　same　time,　TensorRT　technology　is　used　for　reconstruction　and　optimization　to　reduce　the　computation　and　improve　the　detection　speed.　Experimental　results　show　that,　compared　with　YOLOv5,　the　algorithm　at　mAP@.5　and　mAP@.5:.95　increases　1.5%　and　2.3%　respectively,　and　the　detection　frame　number　per　second　increases　150%　to　reach　25FPS,　which　provides　the　possibility　for　real-time　instrument　detection　in　substation　scenarios.　©　2024　SPIE.