Abstract Accurate blueberry ( Vaccinium spp.) detection, maturity classification, and yield estimation are essential for precision orchard management but are hindered by small fruit size, dense clustering, occlusion, and ambiguous maturity color gradients. This study proposes the YOLO‐blueberry detection and classification (YOLO‐BDC) model and dual‐scenario yield estimation approaches to address these challenges. YOLO‐BDC optimizes YOLOv8s with a small‐target detection layer, Spatial Pyramid Pooling with Efficient Layer Aggregation Network feature fusion, efficient channel attention, and transfer learning to enhance generalization. The dual‐scenario yield estimation approaches include regression fitting models and a BP‐Net (Backpropagation Neural Network) model, tailored for canopy‐level and single‐plant‐level yield prediction, respectively. Experiments on a multi‐site dataset show YOLO‐BDC outperforms state‐of‐the‐art models with 81.7% recall and 87.2% mAP50, improving immature and semi‐ripe detection. Experimental results demonstrate that the proposed method achieves an R 2 of 0.97 for counting and 0.93 for maturity classification. Regarding yield estimation, regression models attained an R 2 of 0.93 at the canopy level. Notably, at the single‐plant level, front‐view imagery ( R 2 = 0.93) exhibited superior performance compared to top‐view imagery ( R 2 = 0.84). Furthermore, the BP‐Net model achieved R 2 values of 0.93 and 0.77 for canopy and single‐plant estimation, respectively. This unified technical chain reduces orchard deployment complexity, offering data‐driven support for harvest scheduling and resource allocation to boost blueberry cultivation sustainability and efficiency.
Yang et al. (Sun,) studied this question.