Abstract:
Honey, a valuable and globally consumed food product, has significant market potential linked to its origin. However, authenticating honey is challenging due to sophisticated adulteration techniques. This current research introduces an innovative approach employing YOLOv7, a cutting-edge object detection model, to detect and classify honey pollens, thereby bolstering the authentication of honey. Our methodology involved creating a data set comprising three well-known honey varieties (Sundarban, Litchi, and Mustard), supplemented by three sets of unidentified honey pollen images sourced from Kaggle. Subsequently, we assembled a data set consisting of 3000 images representing the pollen types extracted from the known honey samples. To tackle the challenge of limited sample sizes, we employed data augmentation techniques. The efficacy of our approach was evaluated using established statistical measures including detection accuracy, precision, recall, mAP value, and F1 score, yielding impressive values of 98.3, 99.3, 100, 99.2%, and 0.985, respectively. The YOLOv7 model’s reliability was validated using Kaggle’s unknown honey pollen data sets, which showed that it correctly detected and identified these new pollens based on previous training. Through rigorous experimentation and validation, our study underscores the potential of the YOLOv7 framework in revolutionizing quality control practices within the honey industry, ensuring consumers access to genuine and top-tier honey products through pollen image analysis.
