Abstract: Transmission lines often pass through remote mountainous areas with limited communications, only BeiDou short message communication (SMC) is available for the transmission of monitoring information. Limited by the narrowband and high BER channel characteristics of BeiDou SMC, the monitoring images are not efficiently or steadily transmitted. For this reason, this paper proposes a high-ratio and BER-resistant image compression system adapted to the channel characteristics of BeiDou SMC to realize efficient and reliable monitoring of transmission lines. First, our system designs an adaptive region of interest extraction algorithm to extract key information such as transmission towers and remove redundant pixels. Then, an image coder/decoder based on attention mechanism is built to train lossy data and improve the robustness and BER resistance of the model. The experimental results show our proposed system effectively monitors transmission towers, achieving 97% image compression, and improving the peak signal-to-noise ratio of the reconstructed image by 5.2 dB and the structural similarity by 0.21 compared with the variational autoencoder when BER stands at 5%.