Abstract
Authenticating the botanical origin of honey is crucial for ensuring product integrity, consumer trust, and sustainable trade. However, manual pollen identification under a microscope is time-consuming and prone to human error. To address this, our project developed an AI-based honey authentication pipeline tailored to South African honey, leveraging deep learning for pollen detection, classification, and clustering.
The system uses YOLOv11 to automatically detect and crop pollen grains from microscopy images. Each team member implemented a separate classification pipeline—one using a Convolutional Neural Network (CNN), another a Vision Transformer (ViT), and the third a hybrid CNN–Transformer model. All three classifiers achieved strong performance, with test accuracies above 90% and F1-scores exceeding 0.90, demonstrating robust pollen recognition across architectures.
To handle low-confidence pollen grains, HDBSCAN clustering was applied to feature embeddings This revealed previously unseen pollen groups for expert review. The CNN-based pipeline ultimately provided the most accurate honey authentication results, producing pollen composition reports that most closely matched expert analyses according to ICBB standards.
By integrating detection, classification, and clustering into one automated workflow, our project shows how AI can streamline honey authentication and pollen analysis, supporting apiculture, food traceability, and biodiversity research in the South African context.
Videos 1
Watch presentations, demos, and related content
Like, comment, and subscribe on YouTube to support the creator!
Documents 4
Downloadable resources and documentation
Gallery 1
Explore the visual story of this exhibit
Title Image