Efficient Image Compression with Dual Attention for Signal-Free Transmission Line Surveillance

To enable efficient remote monitoring of transmission lines in areas with limited connectivity, this paper proposes a deep-learning-based image compression method featuring an Efficient Channel-Temporal Attention Module (ETAM). ETAM combines ECA-Net for adaptive channel feature refinement and TAM for capturing temporal dynamics. Integrated within a residual autoencoder and trained with a composite loss function, the model employs context-adaptive entropy coding for further compression. Experiments on the Kodak and STN PLAD datasets show that the method achieves superior performance at 0.10 bpp-PSNR 30.5 dB, SSIM 0.86, LPIPS 0.27-outperforming both traditional and modern codecs. The approach effectively reduces data volume while preserving structural details, making it ideal for low-bandwidth, satellite-based monitoring applications.