Adulteration of honey with inexpensive sweeteners is a growing issue in the honey industry, with an estimated 80% of honey on the market being counterfeit. Rapid, reliable, and cost-effective detection methods are urgently needed to protect both consumers and producers. Thus, this study introduces a novel, non-invasive approach that combines backscattering imaging with deep learning to detect multifloral Heterotrigona itama honey adulteration. A total of 1800 honey speckle images were captured using a backscattering imaging system. The experiment was divided into three parts. In part A, pure honey was adulterated with 1% corn syrup, while in part B, pure honey was blended with corn syrup at varying concentrations of 5, 10, and 20%. Part C involved adulteration using multiple substances, including 30% corn syrup, 15% rice syrup, and 45% apple cider. Three convolutional neural network (CNN) architectures, namely VGG16, InceptionV3, and ResNet34, were trained and evaluated to classify pure and adulterated samples. Among the models tested, InceptionV3 delivered the best performance. It achieved 100% accuracy across all metrics in experiment C and maintained high accuracy even at the minimal adulteration levels of experiment A, which are typically the most challenging to detect. The results demonstrate that the integration of backscattering imaging with InceptionV3 offers a rapid, accurate, and economical method for honey authentication. This technique provides a valuable tool for assessing the purity of high-value honey products and can be further developed for broader applications in food quality control.
Othman et al. (Wed,) studied this question.