Due　to　long-term　exposure　to　harsh　natural　environments,　insulators　are　prone　to　flashover,　breakage　and　self-explosion,　which　can　cause　great　harm,　and　therefore　regular　inspection　of　insulators　is　needed　to　ensure　the　safety　of　the　power　system.　The　traditional　manual　inspection　is　costly　and　inefficient.　To　solve　this　problem,　we　propose　an　insulator　defective　site　detection　algorithm　SR-YOLOv5s　(Structural　Re-parameterization　YOLOv5s)　based　on　improved　YOLOv5s.　First,　we　design　a　Structural　Re-parameterization　Bottleneck　(SRB)　module　and　replace　the　Bottleneck　module　in　the　C3　module　with　the　SRB　module,　which　reduces　the　number　of　parameters　and　GFLOPs　(Giga　FLoating-Point　Operations　Per　Second)　of　the　model.　Then,　we　introduce　Multi-Head　Self-Attention　(MHSA),　which　is　combined　with　the　C3　module　to　construct　a　Cross-Stage　Partial　Bottleneck　Transformer　(CSP　BoT),　which　assists　neural　networks　in　analyzing　the　relationship　between　insulators,　their　defect　features,　and　the　surrounding　environment.　By　assigning　appropriate　weights,　it　enables　more　accurate　identification　of　targets　within　images,　thereby　enhancing　the　model’s　overall　accuracy.　Experimental　results　show　that　the　SR-YOLOv5s　algorithm　proposed　in　this　paper　has　higher　detection　accuracy,　smaller　model　size　and　faster　detection　speed　in　the　defect　detection　task　of　insulators　compared　to　the　original　YOLOv5s　algorithm.　The　method　provides　a　reference　for　defect　detection　of　insulators.　©　The　Author(s)　under　exclusive　licence　to　The　Korean　Institute　of　Electrical　Engineers　2025.