Over the past decade, drones or Unmanned Aerial Vehicles (UAVs) have seen widespread use, and even the government is actively promoting their application. Weed infestation significantly challenges potato cultivation, leading to substantial yield losses. Effective weed management still remains a critical issue for the industry. UAVs with spray systems can apply herbicides directly to weeds, leading to more efficient and targeted treatments. However, for UAV-based weed control, we require accurate data that can later be used for weed detection through deep learning techniques. Once the areas with weed infestations are identified, we can conduct site-specific spraying using UAVs. This study investigates the application of deep learning models, specifically YOLOv8 and YOLOv9, for weed detection in potato fields using UAV-captured aerial images. By training these models on a dataset of annotated images, we aimed to develop a robust system for identifying and locating weeds in potato crops. Both models showed strong classification performance in a dense, weed-infested environment, with YOLOv9 outperforming YOLOv8 in recall (63.4%), mAP@50 (67.5%), and mAP@50–95 (45.6%), indicating superior object detection and localization accuracy. YOLOv8, however, achieved a higher precision (70.5%), minimizing false positives. The findings suggest that YOLOv9’s increased complexity and parameter count allow it to detect objects better, making it more robust for agricultural applications involving weed management. The findings of this research highlight the potential of UAV-based weed management systems powered by deep learning.