Anomaly detection in power transmission infrastructure prevents failures and ensures grid reliability. Recent methods have demon- strated excellent accuracy but struggle with computational efficiency when deployed on resource-constrained devices such as UAVs. Memory-based approaches with dominant performance like PatchCore require external memory banks that significantly increase execution time, while heavyweight normalizing flow models demand substantial computational resources. This paper introduces PowerLiteNet, a lightweight anomaly detection framework that maintains high detection capability while drastically reducing computational requirements. By integrating Squeeze-and-Excitation (SENet) attention with our lightweight architecture, our approach achieves a 57% reduction in computational demands and sub-millisecond inference times (0.6ms versus 179ms). Eval- uated on the Inspection Power Line Asset Dataset (InsPLAD), our SENet-enhanced lightweight model achieved significantly better performance (81.99% mean AUROC) compared to their non-enhanced models (76.69%). This research bridges the gap between advanced deep learning techniques and real-world deployability by balancing computational efficiency with detection accuracy. Our approach offers a scalable automated powerline asset monitoring solution on UAV platforms with limited computational resources.