With　its　safe　and　efficient　characteristics,　the　high-voltage　power　strip　operation　robot　provides　a　new　solution　for　the　maintenance　of　power　equipment　and　promotes　the　intelligent　process　of　power　equipment　maintenance.　In　order　to　solve　the　existing　problem　of　low　efficiency　of　robot　specific　target　perception,　this　paper　proposes　an　improved　YOLACT　model　and　an　application　program　that　combines　the　model　output　with　point　cloud　information.　The　enhanced　YOLACT　model　has　been　improved　in　two　key　aspects　when　compared　to　the　original　model.　First,　a　more　efficient　feature　pyramid　network　is　adopted　to　enhance　the　model＇s　multi-scale　feature　fusion　capability　and　generalization　ability.　Second,　an　attention　mechanism　is　introduced　so　that　the　model　can　better　focus　on　the　sensed　target,　thus　improving　the　accuracy　of　the　model＇s　object　detection　and　segmentation.　The　experimental　results　in　the　self-constructed　dataset　show　that　the　mAP　and　AP90　values　of　the　improved　model　are　higher　than　those　of　the　original　YOLACT　model　by　1.13%　and　2.75%　and　all　other　indexes　are　higher　than　those　of　the　original　model.　The　results　of　combining　the　model　output　with　the　point　cloud　show　that　the　detection　model　can　well　meet　the　requirements　of　specific　target　sensing　for　the　belt　power　operation　robot.　　©　2024　ACM.