Repozytorium publikacji - Politechnika Gdańska

Honeybees plays vital role for the environmental sustainability and overall agricultural economy. Assisting bee colonies within their proper functioning brings the attention of researchers around the world. Electronics systems and machine learning algorithms are being developed for classifying specific undesirable bee behaviors in order to alert about upcoming substantial losses. However, classifiers could be impaired when used for general honeybee colony state inference. Application of the classifier models for the hazardous situations detection without focusing on the model’s genericity could result with systems that are not applicable in the real environment. Furthermore, the detection of a specific phenomenon does not provide researchers with any new conclusions about the honeybee colony life but only with the binary information about hazardous situation presence. In our research we propose a method for inferring the bee colony state using a sensitive contrastive autoencoder and an anomaly detection model. With presented approach, hive’s internal state is modeled with the use of an autoencoder’s latent vector extended with in-hive temperature dynamics. We test our methodology with a bee feeding experiment where the glucose syrup application was detected and the length of food intake was estimated. As our methodology has been applied successfully, we argue that contrastive autoencoders can be used for precise inference about the behavior of honeybees.

Autorzy

• dr inż. Tymoteusz Cejrowski link otwiera się w nowej karcie ,
• dr hab. inż. Julian Szymański link otwiera się w nowej karcie