Advances in agricultural fruit detection using you only look once (YOLO) algorithm: A review

Artificial intelligence-driven, automated, and real-time fruit detection systems are a continuously evolving area of research in agriculture. In recent years, YOLO (You Only Look Once) algorithms have gained popularity and are widely used in agricultural fruit detection tasks. However, fruit detection research is multidisciplinary in nature, and the literature, methods, and algorithms are rapidly evolving, which necessitate thorough documentation, synthesis, and evaluation of the literature. Therefore, this study documents, assesses, and synthesizes the existing literature on YOLO-based object detection through bibliometric analysis (i.e., 620 articles) and systematic literature review (i.e., 114 articles). We observed that the volume of published literature increased by 850% over the span of five years (i.e., 2020 to 2025). YOLOv5 remains the most widely used model (33.3%), while YOLOv8 demonstrated a fast-growing adoption rate (24.6%). Additionally, mean detection performance across studies was 91.1%. Moreover, critical limitations and challenges are also discussed. Only 25.4% of studies reported computational efficiency metrics, while 95% used custom datasets, limiting reproducibility. Major challenges include reduced detection performance for small fruits (5–10% decline), severe occlusions, and accuracy-speed trade-offs. Priority research directions include hardware-aware optimization, standardized benchmark development, and domain adaptation for cross-orchard generalization. These insights guide the advancement of YOLO-based fruit detection from laboratory studies toward reliable agricultural deployment.