Insulator shackles are critical metal connectors in outdoor transmission lines. Long-term exposure to harsh environments makes them prone to corrosion, which may lead to power grid safety incidents. Meanwhile, the complexity of natural environments further increases the difficulty of manual inspection and assessment. To address this issue, this paper proposes an improved small-object detection model based on YOLOv8, named YOLO-Improved. The model extends an additional detection head branch to enhance detection sensitivity for small-area corrosion features. Furthermore, the model integrates the omni-dimensional dynamic convolution (ODConv) and the C2f bitemporal feature aggregation module (C2f_BFAM) to construct a plug-and-play module, referred to as OCB (ODConv and C2f_BFAM), which is embedded into the neck network. Specifically, ODConv adaptively adjusts convolution kernels across four dimensions: the number of kernels, spatial size, input channels, and output channels, thereby strengthening the multi-scale feature representation capability for insulator shackles. Meanwhile, C2f_BFAM adopts a parallel branch processing and feature fusion structure to effectively integrate shallow detailed features with deep semantic features, thereby improving feature discrimination capability. Experimental results on an insulator shackle corrosion dataset show that, YOLO- Improved significantly improves detection accuracy and reduces the missed detection rate compared with the baseline YOLOv8 model, and also outperforms the current state-of-the-art YOLOv26 model. Furthermore, the proposed model demonstrates good generalization ability and a lower missed rate on both the same-domain PTL-AI Furnas insulator defect dataset and the cross-domain COCO dataset.