Improved YOLOv8 Algorithms for Agricultural Monitoring and Harvesting Tasks: A Comprehensive Review

Accuracy and real-time performance are two major challenges in monitoring plant growth, detecting crops, and recognizing diseases in complex real-world agricultural environments. Growing environments present significant difficulties for object detection due to factors such as variable weather and lighting conditions, shooting distances, varying degrees of occlusion, and diverse morphological characteristics. The YOLO series of models, especially the prominent YOLOv8, are state-of-the-art models for object recognition that have revolutionized the field by achieving an optimal balance between speed and accuracy. Since YOLOv8 appeared two and a half years ago, many improvement measures or modifications have been proposed in the literature for different detection tasks and applications. This paper systematically reviews these Improved YOLOv8 algorithms, focusing on object detection in plants (e.g., crops, diseases, and growth stages), to evaluate the proposed changes or improvements. Inspired by the reviewed architectures and comparative analyses, we propose a modular architecture called PLANT-YOLOv8 based on the YOLOv8 framework. The proposed modular configuration of the YOLOv8 structure is flexible, easy to implement, and extendable. Additionally, our analysis provides recommendations and potential improvements for each YOLOv8 component that could be replaced or enhanced. Lastly, we present and evaluate Improved YOLOv8 architectures from the reviewed literature to demonstrate their composition and complexity as prime examples of our modular PLANT-YOLOv8 architecture.