Using deep learning approaches to detect cells in honeycombs

Harvesting the honeycomb using the right methods is of great importance for the beekeeping activities and for the country's economy. Failure to use the right methods in honey harvest will negatively affect both the production quantity and quality of honey. Unconscious beekeeping activities prevent bees, which have a great role in the continuation of natural life, from continuing their own generation. In this thesis, it is mainly mentioned about beekeeping, beehives and honeycombs, the approach of obtaining images in a honeycomb and labeling these images. The methods and evaluation metrics that can be used for the classification of the obtained labeled images are examined. In addition, in this thesis study, a honeycomb data set containing a total of 103,451 training and 25,863 test images, consisting of 7 classes, was created as an example. Different deep learning algorithms (VGG16, VGG19, Inception-V3, AlexNet, SSCNN and ResNet50V2+Xception) were used for classification of honeycomb images. The success of deep learning approaches can vary according to the problem. Therefore, these deep learning architectures were run on the same dataset and compared according to the experimentally obtained results. As a result of the classification process, the highest success rate was obtained from AlexNet deep learning algorithm with 95%.