Algorithm operation steps.

The detection of insulator defects in transmission lines is of paramount importance for the safe operation of power systems. However, small object detection faces numerous challenges, such as significant difficulty, substantial interference from complex backgrounds, and inconsistent annotation quality. These factors continue to constrain the performance of existing methods. To address these issues, this paper proposes an improved object detection algorithm named SCI-YOLO11, which optimizes the YOLO11 framework from three aspects: feature extraction, attention mechanism, and loss function. Specifically, to tackle the difficulties associated with small object detection, we replace conventional convolutions in the Backbone with SPDConv modules to enhance feature capture capabilities for small targets and low-resolution images while reducing computational overhead. To improve model accuracy further, we introduce the SE attention mechanism that adaptively adjusts the weights of feature channels to enhance the discriminative ability of insulator defect features. In response to the adverse effects caused by inconsistent annotation quality on defect image detection performance, we incorporate Wise-IoU-V3 loss function to optimize boundary box regression performance effectively mitigating negative impacts stemming from uneven annotation quality. Experimental results demonstrate that SCI-YOLO11 achieves a 3.2% improvement over baseline models in terms of MAP@0.5 metric; precision and recall rates increase by 2.6% and 3.7%, respectively. Additionally, its parameter count and floating-point operations decrease by 6% and 7.9%, respectively. These experimental findings substantiate the substantial improvements in detection accuracy, lightweight design, and robustness provided by SCI-YOLO11. This framework offers an effective technical solution for identifying defects in transmission line insulators.