Abstract: With the development and maturity of computer vision technology, visual sensors have become standard equipment for various types of UAVs. Introducing visual navigation assistance on the basis of traditional inertial/GNSS integrated navigation, and obtaining carrier pose information from the rich environmental information and carrier motion information provided by visual images, can effectively solve the problem of high-precision navigation in GNSS-denied scenarios. A novel multimodal navigation enhancement technology integrating vision, IMU and GIS map is proposed in this paper. The AI model for heterogeneous image matching (SuperPoint+SuperGlue) is used to achieve accurate matching between real-time aerial images and reference satellite images, and Kalman filtering is combined to complete multimodal data fusion. Experimental results show that in low-altitude environments of 100~1 000 m with GNSS signal failure, the system positioning accuracy error is less than 5 m (RMSE).