Project DetailsIn this project, we developed an automated system for detecting and localizing Ziziphus pollen grains in microscopic images of honey to determine their relative proportion among all grains in the image. This allows us to deduce the primary composition of each honey sample and decide on its botanical classification as monofloral or polyfloral. Following an extensive literature review, we selected the YOLOv8 architecture to implement our solution. Various experiments were conducted, primarily focused on data preprocessing and initial training phases, to enhance model performance. The project's main innovation was the creation of a new dataset based on the original dataset, which led to improved Ziziphus grain classification capabilities, despite annotation challenges encountered during the work. Ultimately, a fine-tuning process resulted in an 84% accuracy on the test set. Additionally, we built a grain-counting system that, according to a statistical assessment, achieves a detection error rate of approximately 5%. We believe this project may impact the honey industry and could potentially be expanded to classify and localize a broader range of pollen grains, ultimately leading to a fully autonomous system.
