Automatic Detection of Honey in Hive Frames using Deep Learning

In recent years, smart technology has become increasingly useful for monitoring honeybee colonies' health and condition in real time using a remote monitoring system. Due to the development of new technologies, it is possible to utilize deep learning techniques in order to improve the understanding of honey conditions within hives. In this study, we propose a method for automatic honey detection in honeycomb frames. A dataset of images of hive frames was collected and annotated by experts. We employed transfer learning by fine-tuning several pre-trained convolutional neural network (CNN) architectures using the image dataset. The best-performing image classification model was VVG19 with an accuracy of 84% and an F1-score of 84% on the test set. As demonstrated in this study, transfer learning can be a useful method of analysing images remotely without human intervention or physical access to remote beehives. Manpower requirements could be reduced and productivity could be improved, particularly in rural areas.