As critical components of transmission lines

insulators require reliable anomaly detection to ensure the safe and stable operation of power systems. Existing zero-shot anomaly detection methods face challenges in insulator inspection scenarios

including complex background interference and disturbances caused by reflections

shadows

and connecting fittings on insulators. Moreover

existing public datasets lack specialized data for insulator inspection scenarios. To address these issues

this paper constructs an insulator anomaly test dataset collected from on-site unmanned aerial vehicle(UAV) inspections of transmission lines

and proposes a zero-shot insulator anomaly detection method integrating saliency guidance and feature enhancement

named saliency-guided and feature-enhanced CLIP(SAFE-CLIP). To address complex background interference

a saliency-guided dynamic background suppression strategy is proposed to effectively suppress background noise while preserving the distribution structure of foreground anomaly scores. To address the interference caused by reflections

shadows

and connecting fittings on insulators

an enhanced content-guided attention(ECGA) module is proposed

which fuses attention information across spatial

channel

and pixel dimensions

moreover

residual connections are introduced to protect the feature space of pre-trained CLIP

significantly reducing false positive detection on normal foreground regions. Experimental results demonstrate that the proposed method achieves a pixel-level average precision of 56.07% while maintaining high image-level detection rates

which is increased by 5.83% over the baseline model

and significantly outperforms existing state-of-the-art methods.