.png){kind=icon}
.png){kind=icon}
Return
NECTAR: novel ensemble comb targeting & automated reporting
.png){kind=icon}
PRE
AI
E
A
DOI：10.1016/j.knosys.2026.116801.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Honeybees are crucial to ecosystem continuity and global food production through pollination, making reliable colony health monitoring essential for sustainable agriculture. However, inspections are still largely manual, while the fine-grained multi-class structure of honeycomb cells, severe class imbalance, tiny objects, and biologically similar categories make automated analysis challenging. Manual inspections are labor-intensive, prone to human bias, difficult to sustain regularly, and unable to support large-scale, data-driven monitoring; in contrast, automated analysis has the potential to accelerate decision-making, reduce operational costs, and positively impact production efficiency and commercial sustainability. In this study, we propose NECTAR (Novel Ensemble Comb Targeting & Automated Reporting), a comprehensive honeycomb-cell analysis framework. A newly curated, high-resolution, multi-class dataset was created. Modern object detectors—YOLO-based CNNs and RT-DETR—were systematically evaluated under identical conditions. Subsequently, ensemble strategies were applied, yielding clear gains—particularly in rare and biologically critical classes. The proposed approach achieved up to 0.90+ AP@0.5 and 0.85+ AP@[0.50:0.95]. In addition, NECTAR presents one of the most detailed literature syntheses in this domain, structuring existing research and clearly identifying the methodological gap addressed by this work. Beyond detection accuracy, an LLM-based reporting layer was integrated to convert numerical outputs into interpretable, rule-guided colony health summaries, providing preliminary decision-support capabilities for beekeeping applications. Overall, NECTAR presents one of the most comprehensive frameworks in the literature by combining a new dataset, multi-model comparative analysis, ensemble fusion, and LLM-supported interpretation, offering a robust and promising contribution toward next-generation smart apiculture.
Keywords:
beekeeping
honeybee comb detection
deep learning
ensembles
LLM
Journal
K
IF:
7.6
Papers:
1.2W
Citations:
4.5W
Organization
Cited Papers