Conventional monitoring of oil palm fresh fruit bunches (FFBs) relies heavily on manual inspection, which is labour-intensive and often fails to deliver timely information on fruit maturity and yield potential required for precision harvesting. This study proposes a drone-based, real-time FFB classification system using deep learning to support efficient plantation monitoring, harvesting management, and yield estimation. A custom dataset comprising 11,613 annotated images of oil palm trees with visible FFBs was developed, categorising bunches into six distinct ripeness classes: Abnormal, empty bunch, underripe, ripe, unripe, and overripe. The dataset was divided into training (9,121 images), validation (1,657 images), and testing (835 images) subsets. Object detection models based on the Ultralytics YOLO11 architecture were trained for 100 and 200 epochs and deployed within a Python application developed using Streamlit, OpenCV, and the Ultralytics API. The system processes live unmanned aerial vehicle (UAV) video streams, performs frame-level FFB detection, tracks FFB across consecutive frames, and records per-tree detection results in comma-separated values (CSV) format using QR-code-based tree identification. Experimental evaluation shows that the YOLO11 model achieved a mean average precision (mAP) of 98.3%, precision of 94.0%, recall of 97.2%, and fast inference speed of approximately 7.7 ms/frame. The results demonstrate that the proposed system offers a scalable and practical solution for automated, real-time FFB monitoring, supporting ripeness-based harvest planning and operational efficiency in oil palm plantations.