Deep learning for horticultural innovation: YOLOv12s revolutionizes micropropagated lingonberry phenotyping through unified phenomic-genomic-epigenomic detection

• Phenomic analysis of micropropagated plants (1.0–99.5% precision across growth environments) • Culture vessel identification (79.6–89.5% accuracy) • Genomic-epigenomic marker detection (32.2–74% ladder precision) • The model's computational efficiency (21.2 GFLOPs) enables real-time processing (10.6–11.3 ms/frame) on mobile devices • 40% faster accurate trait selection compare to YOLOv8 and YOLOv9 model • Standardized quality control for plant tissue culture industry • Greenhouse-deployable yield prediction (67.3% mAP50) Vaccinium vitis-idaea L. (lingonberry), globally recognized as a superfruit for its medicinal properties, has long been cultivated and consumed by Canadian Indigenous communities. This study introduces an AI-powered surveillance system that leverages an optimized You Only Look Once (YOLOv12) architecture to revolutionize yield estimation, phenomic profiling, and genomic/epigenomic analysis in micropropagated lingonberry. A custom multi-class annotated dataset was developed to evaluate model performance under real-world conditions. The YOLOv12 model, built on a RELAN backbone with flash-attention mechanisms, excelled in global context modeling, enabling accurate detection of berries and regenerated shoots in both ex vitro and in vitro environments. In contrast, YOLOv8 and YOLOv9, which rely on CNN-based feature extraction, demonstrated computational efficiency but suffered from overfitting and reduced operational robustness. In multi-class detection scenarios, YOLOv12 achieved the highest mean Average Precision with 67.3% mAP@50 in yield detection, 1.0–99.5% mAP@50 in micropropagated plant trait detection (shoots, berries, flowers), and 32.2–74% accuracy in gel electrophoresis band detection. These results reflect a 22% increase in throughput and a 38% reduction in error rates compared to conventionally human-monitored methods, significantly reducing labor cost for plant breeders and agricultural biotechnologist. The integrated system enables simultaneous monitoring of phenotypic traits across growth stages and precise molecular band analysis, establishing a new paradigm for precision agriculture and lingonberry improvement. This work establishes YOLOv12 as the first unified framework for micropropagated lingonberry phenotyping across biological scales, demonstrating labor reduction in breeding programs while maintaining operational reliability. The technology's mobile compatibility and cloud-integration potential offer immediate applications for the global $2.3B lingonberry market, particularly in precision nurseries and nutraceutical production.