Abstract: High-precision positioning and navigation are among the key technologies for achieving autonomous flight in unmanned aerial vehicles (UAVs). Image matching-based visual navigation technology, characterized by its simple device structure and high passive positioning accuracy, can be integrated with inertial systems to form a highly autonomous high-precision navigation system. The core of the image matching system lies in the registration of real-time captured images with pre-stored reference images. Due to the high-speed flight of UAVs and the multi-source nature of the reference images, this poses high demands on the accuracy, speed, and robustness of image registration. To address this issue, we propose a descriptor that is both high in accuracy and robustness, and facilitates image matching. This descriptor, termed DSOG, describes the pixel features of the image orientation's gradients. DSOG possesses outstanding image matching performance and computational efficiency. Additionally, an optimized similarity measurement matching template is proposed, which refines the traditional fast similarity measurement algorithm based on feature representation using Fast Fourier Transform (FFT) in the frequency domain, thereby eliminating redundancy in the matching process. The matching framework proposed in this paper has been evaluated with numerous different types of multimodal images, and the results indicate that it outperforms state-of-the-art methods in terms of matching performance.