Machine　learning　based　on　hand-crafted　features　from　the　original　aerial　insulator　image　has　shown　promising　results　in　the　insulator　defect　detection　in　recent　years.　Such　methodologies,　nevertheless,　are　not　suitable　for　detecting　insulator　defect　in　complex　backgrounds,　which　intrinsically　relies　on　sufficient　prior　knowledge,　low　background　interference,　and　certain　object　scales.　Therefore,　in　this　study,　an　effective　insulator　defect　detection　model　using　improved　Faster　Region-based　Convolutional　neural　network　(Faster　R-CNN)　based　on　the　original　aerial　insulator　image　is　proposed.　The　soft　non-maximum　suppression　(Soft-NMS)　is　utilized　to　improve　the　performance　of　detection　overlap　insulator　and　ResNet　101　model　is　adopted　to　effectively　extract　feature　from　insulator　images.　Furthermore,　the　proposed　method　has　more　accurate　target　location　and　higher　average　accuracy　by　comparing　traditional　Faster　R-CNN.　©　2019　IEEE.