In recent years, the application fields of UAV technology have been expanding, covering a wide range of fields such as military and civilian. Especially in the field of absolute visual localization, the introduction of deep learning has brought significant breakthroughs to traditional methods and promoted the technological development in this field. Compared with the early methods relying on traditional computer vision feature extraction, absolute visual localization methods based on deep learning have achieved significant improvements in localization accuracy and robustness. In this paper, we systematically review deep learning-based absolute visual localization methods for UAVs, and explore the application potential and development prospects of absolute visual localization technology by analyzing the limitations of global navigation satellite systems and relative visual localization in complex environments. The key research progress in this field is summarized, including dataset construction and deep learning model optimization, and the current technical bottlenecks are dissected. Finally, future research directions are proposed to further promote the wide application and sustainable development of absolute visual localization technology in the field of autonomous UAV navigation.