YOLO11-ALi: an improved YOLO11-based model for blueberry target detection from full flowering to fruit expansion in greenhouse environments.

The abscission of blueberry floral organs after fruit set is crucial for fruit development and quality, while the closed, high-humidity greenhouse environment often causes withered flower retention, adversely affecting blueberry growth and commercial value. Current blueberry detection technologies focus on maturity classification, health monitoring and automatic picking, leaving a gap in complex target detection during the full flowering to fruit expansion period. To address this, this study proposes YOLO11-ALi, a modified model based on YOLO11, for greenhouse pot-grown northern highbush blueberries. Three key improvements are made to the baseline: replacing the SPPF module with AIFI to reduce small-target detail loss, integrating LSKAttention into C2PSA to form C2PSA_LSKA for enhanced small-target attention, and incorporating iAFF into the neck C3k2 module to improve cross-layer feature fusion efficiency. A dedicated dataset of 4527 images covering full flowering to fruit expansion was constructed, and ablation experiments verified the modules' roles and synergistic effects. Results show that the integrated model achieves optimal performance: compared with the original YOLO11, precision, recall, mAP@50 and mAP@50:95 are increased by 0.6%, 1.8%, 1.2% and 6.8% respectively, with computational complexity and inference speed meeting real-time field detection requirements. The proposed model provides a reliable technical basis for intelligent monitoring and refined management of greenhouse blueberries.