Transmission line operations are inherently high-risk and require real-time safety monitoring and control to promptly detect and correct anomalies.Existing intelligent monitoring solutions suffer from limitations such as insufficient modality coverage

delayed response times

and poor adaptability to edge computing environments.To address these challenges

this paper proposes a lightweight multimodal fusion framework.The system achieves an inference latency of 43 ms on the NVIDIA Jetson AGX Xavier edge device—an 83.5% reduction compared to full-precision models.Model parameters are compressed to 2.8 MB(a 98.7% reduction)

and memory usage is reduced by 86.9%

meeting the demands of real-time response with millisecond precision.