YOLOv8-based on-the-fly classifier system for pollen analysis of Guindo Santo (Eucryphia glutinosa) honey and assessment of its monoflorality

The unique characteristics of Chilean flora make it a subject of valuable research into honey produced by endemic species, which holds significant cultural and economic importance, particularly for local ethnic groups. A prime example is the exploration of honey from the recently discovered Guindo Santo ( Eucryphia glutinosa ) species in Alto Bio Bío, Chile. This area is culturally and economically tied to the Pehuenches community, which is part of the Mapuche culture and resides in the south-central Andes. Currently, one challenge facing palynology worldwide is automating pollen counting, which is crucial for determining the floral origin of honeys. We propose leveraging digital object detection tools and artificial intelligence to achieve this automation. Our study employs bright-field microscopy with YOLOv8 on Google Colab, a state-of-the-art neural network for object detection, to count pollen and ascertain the floral origin of Guindo Santo honey. By training the network with images of typical pollen grains from Guindo Santo honey, our auto-validation test yielded a mean average precision (mAP) of 97.6 % for all classes and 94.4 % for Guindo Santo. Both automatic and manual techniques confirm the monofloral nature of Guindo Santo honey, albeit with some discrepancies in percentages. We analyze this in terms of experimental methods and propose solutions to address it. • High Detection Accuracy: The YOLOv8 model achieved a mean average precision (mAP) of 97.6 % in pollen grain classification, demonstrating its effectiveness in automated pollen identification in Guindo Santo honey. • Cultural and Economic Importance: Guindo Santo honey is crucial for the Pehuenche community in Chile, but the lack of scientific validation regarding its botanical origin limits its market value. • Automation of Pollen Counting: The research proposes an automated system that overcomes the limitations of traditional pollen counting methods, improving the speed and accuracy of the process. • Development of a Pollen Database: A pollen database was created from local species, serving as the foundation for training the YOLOv8 model, facilitating precise identification in honey analysis.