Classification of Subspecies of Honey Bees using Convolutional Neural Network

Honey bees are vital for humans and the ecosystem; humans use bee products for various purposes and honey bees play an essential role in pollination. Identifying the subspecies of honey bees is essential for beekeepers, bee breeders, and researchers. Conventional methods of identification are time-consuming and expensive, as such, a more efficient and accurate technique is required. This paper proposes a computer vision-based method to classify subspecies of honey bees, with a dataset consisting of seven honey bee subspecies utilised. Due to the unavailability of a sufficient amount of labelled images to train a model, Transfer Learning (TL) technique has been employed. The Inception v3 model has been utilized as the main Convolutional Neural Network architecture; trained initially using the ImageNet dataset and then fine-tuned using the honey bee dataset. Image augmentation technique has also been used to reduce the impact of an imbalanced and small dataset. The Transfer Learning-based Inception v3 model was trained using the Sigmoid Focal loss function and demonstrated an accuracy of 94% with a weighted average F1 score of 0.94. This demonstrates the applicability of computer vision in the classification of honey bees.