Real-Time On-the-Go Annotation Framework Using YOLO for Automated Dataset Generation

Accepted for publication in the Proceedings of the 5. Interdisciplinary Conference on Electrics and Computer (INTCEC 2025) 15-16 September 2025, Chicago-USA Real-Time On-the-Go Annotation Framework Using YOLO for Automated Dataset Generation Mohamed Abdallah Salem dept. Agricultural and Biosystems Engineering North Dakota State University Fargo, USA Mohamed.Salem@ndsu.edu Ahmed Harb Rabia dept. Agricultural and Biosystems Engineering North Dakota State University Fargo, USA Ahmed.Rabia@ndsu.edu Abstract—Efficient and accurate annotation of datasets re- mains a significant challenge for deploying object detection models such as You Only Look Once (YOLO) in real-world applications, particularly in agriculture where rapid decision- making is critical. Traditional annotation techniques are labor- intensive, requiring extensive manual labeling post data collec- tion. This paper presents a novel real-time annotation approach leveraging YOLO models deployed on edge devices, enabling immediate labeling during image capture. To comprehensively evaluate the efficiency and accuracy of our proposed system, we conducted an extensive comparative analysis using three prominent YOLO architectures (YOLOv5, YOLOv8, YOLOv12) under various configurations: single-class versus multi-class annotation and pretrained versus scratch-based training. Our analysis includes detailed statistical tests and learning dynamics, demonstrating significant advantages of pretrained and single- class configurations in terms of model convergence, performance, and robustness. Results strongly validate the feasibility and effectiveness of our real-time annotation framework, highlighting its capability to drastically reduce dataset preparation time while maintaining high annotation quality. Index Terms—Autonomous control, greenhouse automation, light intensity regulation, precision agriculture, Q-learning, re- inforcement learning, temperature control I. INTRODUCTION The advent of precision agriculture has underscored the necessity for rapid, accurate, and scalable methods to monitor and manage crop health and weed proliferation. Traditional manual annotation of agricultural datasets is labor-intensive, time-consuming, and prone to human error, thereby impeding the timely deployment of machine learning models in dynamic field conditions [1]. As the agricultural sector increasingly adopts automation and artificial intelligence (AI) technologies, there is a pressing demand for efficient annotation frameworks that can operate in real-time, facilitating immediate data label- ing and model refinement [2]. Object detection models, particularly those based on the You Only Look Once (YOLO) architecture, have demonstrated This research is based upon work supported by North Dakota State University and the U. S. Department of Agri- culture, Agricultural Research Service, under agreement No. 58-6064-3-011. remarkable efficacy in various agricultural applications, in- cluding pest detection, crop monitoring, and yield estima- tion [3], [4]. YOLO’s real-time processing capabilities and high accuracy make it an ideal candidate for deployment in agricultural settings where rapid decision-making is crucial [5]. However, the effectiveness of these models is heavily contingent upon the quality and quantity of annotated data, which remains a significant bottleneck in the development pipeline [6] To address this challenge, we propose a novel real-time annotation framework that leverages YOLO models deployed on edge devices to facilitate immediate labeling during image capture. This approach aims to streamline the data annotation process, reduce latency between data collection and model training, and enhance the overall efficiency of deploying AI models in agricultural environments. By integrating real-time detection and annotation, our framework seeks to empower agricultural practitioners with tools that can adapt swiftly to the evolving conditions of the field. This paper introduces an on-the-go annotation framework that integrates YOLO-based real-time object detection with live manual class labeling and YOLO-format export. Our contributions include: • A lightweight annotation pipeline deployable on edge devices. • A live labeling interface for in-field selective annotation. • Evaluation of 12 YOLO model training configurations. • Statistical analysis comparing single vs. multi-class and pretrained vs. scratch training. II. RELATED WORK The integration of deep learning techniques in agriculture has witnessed substantial growth, with object detection mod- els like YOLO playing a pivotal role in advancing preci- sion farming practices. YOLO’s ability to perform real-time object detection has been harnessed in various agricultural applications, such as identifying pests on crops, monitoring plant health, and facilitating automated harvesting [3], [7]–[9]. arXiv:2512.01165v1 [cs.CV] 1 Dec 2025 Fig. 1. Overview of the proposed real-time YOLO annotation framework. These app

This content is AI-processed based on open access ArXiv data.