Related Experiment Video
Updated: May 20, 2025
08:11
Collection and Identification of Pollen from Honey Bee Colonies
Published on: January 19, 2021
7.0K
Digital image processing combined with machine learning: A novel approach for bee pollen classification.
Caroline Simão1, José Elton de Melo Nascimento2, Vagner de Alencar Arnaut de Toledo2
1State University of Midwestern at Paraná (UNICENTRO/CEDETEG), Alameda Élio Antonio Dalla Vecchia, Vila Carli, 85040-167 Guarapuava City, Paraná, Brazil.
Food Research International (Ottawa, Ont.)
|May 17, 2025
Summary
This study uses digital image processing and machine learning to classify bee pollen, achieving 98.9% accuracy. This method offers an efficient way to ensure the authenticity and quality of bee pollen products.
Area of Science:
- Apiculture and Pollen Analysis
- Machine Learning Applications
- Digital Image Processing
Background:
- Bee pollen classification is vital for product authenticity, quality control, and fraud prevention, especially for high-value stingless bee pot-pollen.
- Traditional pollen analysis methods are often time-consuming and complex, necessitating more efficient alternatives.
Purpose of the Study:
- To develop and evaluate an efficient method for classifying pollen loads from Apis mellifera and pot-pollen from stingless bee species.
- To leverage digital image processing and machine learning for pollen classification based on color patterns.
Main Methods:
- Collected 246 pollen and pot-pollen samples from five bee species.
- Captured high-resolution images using a smartphone and extracted color parameters (R, B, H, V).
- Tested classification models including CatBoost, XGBoost, Random Forest, and k-Nearest Neighbors (kNN); validated with Linear Discriminant Analysis (LDA).
Main Results:
- The CatBoost model achieved the highest performance, with 100% accuracy in training and 98.9% in testing.
- XGBoost, Random Forest, and kNN achieved testing accuracies of 77%, 78%, and 78%, respectively.
- LDA successfully grouped pollen samples into five distinct clusters corresponding to the bee species.
Conclusions:
- Combining digital image processing with machine learning provides an effective and efficient approach for classifying pollen from different bee species.
- This technique shows promising applications in apiculture for ensuring bee pollen product quality and authenticity.
