Spacers in power transmission lines are primarily employed to maintain an appropriate distance between conductors, preventing collisions or contact between them. As the scale of power transmission networks continues to expand, the detection of spacer bars on transmission lines becomes critical for ensuring the safety of the conductors. To address these issues, this paper proposes an improved spacer detection algorithm based on YOLOv12. First, to mitigate the impact of various natural environmental factors that often degrade image quality, an image preprocessing strategy is introduced to enhance the clarity of training samples. Second, to address the small proportion of spacers in the image and the risk of losing crucial information during convolutional down-sampling, the Space-to-Depth Convolution (SPDConv) block is incorporated to preserve informative features and improve detection accuracy. Additionally, due to the lack of publicly available datasets tailored to this task, a new spacer image dataset covering a range of complex real-world scenarios was constructed. Experimental results on this dataset demonstrate that the proposed model outperforms the baseline YOLOv12 model, achieving an 8.2% improvement in precision and a 10.8% increase in recall. These results confirm the effectiveness and feasibility of the proposed approach for accurate spacer detection under complex real-world conditions.