Abstract: The significant improvement of BDS-3 short message communication capability provides users in navigation and other industries with the possibility of narrow-band non real-time image data transmission. According to the characteristics of BDS-3 short message, this paper designs a very low bit rate compression method of deep learning navigation image using generative countermeasure network which introduce a hyperprior to fuse encoder and decoder and establishes navigation image data sets covering different ship types, accident types and meteorological conditions for training and testing. The experimental results show that: compared with JPEG and JPEG2000 and bmshj2018, this method can maintain good visual characteristics under high compression rate, and can realize the transmission of 320×240 size images in a short message. The method has been verified on BDS-3 short message data transmission terminal and Beidou sorting platform, and has good application value in marine support and maritime management.