The paper gives an extensive review of deep learning-based techniques for vision-guided UAV navigation. Conventional UAV navigation methods depend on LiDAR, GPS and IMUs which face challenges in GPS denied environments and are also costly. Vision-based navigation has emerged as a viable solution that uses onboard cameras and computer vision algorithms. Deep learning enhances navigation by enabling tasks such as object detection, scene interpretation and feature extraction. The paper reviews various DRL techniques for UAV navigation highlighting their application in complex environments. It also explores CNN-based methods for autonomous path planning, terrain understanding, and sensor fusion. Applications of deep learning in UAV navigation are presented and also key challenges including data dependency, real-time processing constraints, and the need for safety and reliability are also discussed. The paper suggests future research directions such as deep SLAM, energy-efficient navigation, and cross-domain applications. Emphasis is placed on improving real-time processing through model compression techniques and enhancing standardization through the development of publicly available datasets and benchmarks. Overall, the review intends to offer insights into the advancements and challenges of deep learning integration in UAV navigation, promoting the evolution of more sophisticated and adaptable navigation systems for different applications.