The autonomous decision-making capability of spacecraft is crucial for future missions, allowing for more versatile, independent, and affordable operations in space. The Earth-Moon system (i.e., cislunar space) is expected to become strategically important as national space agencies and private companies leverage the Moon as a platform to reach farther destinations. With increased traffic in the Earth-Moon region, a spacecraft's ability to autonomously navigate will be essential for the safer and more sustainable exploration of space. The objective of this investigation is to assess the feasibility of efficient autonomous optical navigation in the vicinity of the Moon. This Thesis implements recent breakthroughs in deep learning, leveraging convolutional neural networks (CNN), for computer vision. In the case of a spacecraft traveling to the Moon, the CNN is trained to identify key surface features, such as craters, to assist with the autonomous-relative-navigation process. Though CNNs have been proven reliable in terrain-relative navigation (TRN) close to the lunar surface, their applicability at extended distances remains unexplored. This study present a novel approach that harnesses CNNs for crater detection up to 70000 km using synthetic images with annotated craters from a comprehensive database. We employ the computer-vision model YOLOv5 for object detection and utilize POV-Ray for realistic lunar-surface rendering, along with JPL SPICE for precise Sun and spacecraft positioning. Of the three models trained, the model trained with the 10000 largest craters on the surface of the Moon displayed the highest potential for generalized crater detection. Key factors influencing the area and angular predictions were analyzed to inform on the feasibility and expected performance of autonomous navigation. Our findings display an average crater area percent error of 6.834% and an average normalized angular error of 0.088%. Post-analysis of our results indicate that the parameters affecting the predictions the most include: crater radii, crater visibility, radial distance, spacecraft-incidence angle, and solar-incidence angle. This suggests that, with refinement to the training model, a CNN-based, crater-detection scheme is feasible for autonomous optical navigation, with performance metrics suitable for practical applications. This underscores the potential of advanced machine-learning techniques in supporting the future of autonomous space exploration, contributing to the goal of sustainable and autonomous operations in cislunar space. Future work should focus on reducing the variability of the model's predictions to further enhance reliability and accuracy under diverse lunar conditions.