Maturity detection and counting of blueberries in real orchards using a novel STF-YOLO model integrated with ByteTrack algorithm

作者：Na Wu, Jie Wu, Zhenzhen Wang, Yun Zhao, Xing Xu, Yali Wang, Petr Skobelev, Yanan Mi · 发表于：Frontiers in Plant Science · 年份：2025 · DOI：10.3389/fpls.2025.1682024 · 被引用次数：5 · 研究领域：Smart Agriculture and AI、Spectroscopy and Chemometric Analyses、Advanced Chemical Sensor Technologies

Introduction: Blueberries are highly prized for their nutritional value and economic importance. However, their small size, dense clustering, and brief ripening period make them difficult to harvest efficiently. Manual picking is costly and error-prone, so there is an urgent need for automated, high-precision solutions in real orchards. Methods: We proposed an integrated framework that combined the STF-YOLO model with the ByteTrack algorithm to detect blueberry maturity and perform counting. Together with ByteTrack, it provided consistent fruit counts in video streams. STF-YOLO replaced the YOLOv8 C2f block with a Detail Situational Awareness Attention (DSAA) module to enable more precise discrimination of maturity. It also incorporated an Adaptive Edge Fusion (AEF) neck to enhance edge cues under leaf occlusion and a Multi-scale Neck Structure (MNS) to aggregate richer context. Additionally, it adopted a Shared Differential Convolution Head (SDCH) to reduce parameters while preserving accuracy. Results: On our orchard dataset, the model achieved an mAP50 of 79.7%, representing a 3.5% improvement over YOLOv8. When combined with ByteTrack, it attained an average counting accuracy of 72.49% across blue, purple, and green maturity classes in video sequences. Cross-dataset tests further confirmed its robustness. On the MegaFruit benchmark (close-range images), STF-YOLO achieved the highest mAP50 for peaches (91.6%), strawberries (70.5%), and blueberries (90.6%). On the heterogene...