TY - GEN
T1 - WE-Bee: Weight Estimator for Beehives Using Deep Learning
AU - Anwar, Omar
AU - Keating, Adrian
AU - Cardell-Oliver, Rachel
AU - Datta, Amitava
AU - Putrino, Gino
PY - 2022/2/28
Y1 - 2022/2/28
N2 - We present WE-Bee, a hybrid for soft sensing and time series forecasting, to estimate the daily weight variations of honeybee hives. Weight variations of a honeybee hive are the most important indicator of hive productivity, and the health and strength of a bee colony. Precise measurement of the weight of a hive requires an expensive weighing scale under each hive. On the other hand, sensors deployed inside the hive are cheaper than a weighing scale, and are shielded from the extreme weather variations outside the hive. In this work, honeybee activity is monitored using data from sensors inside the hive, along with monitoring the information related to the seasons, time of the day, external weather and the size of hive. WE-Bee's deep learning algorithm is based on Bidirectional LSTM encoder-decoder and attention mechanism. The results from field deployments demonstrate that the system is capable of cumulative weight estimation over multiple weeks. The results also show that the use of appropriate features and the diversity of training data is essential for robust performance of deep learning in the wild for this application.
AB - We present WE-Bee, a hybrid for soft sensing and time series forecasting, to estimate the daily weight variations of honeybee hives. Weight variations of a honeybee hive are the most important indicator of hive productivity, and the health and strength of a bee colony. Precise measurement of the weight of a hive requires an expensive weighing scale under each hive. On the other hand, sensors deployed inside the hive are cheaper than a weighing scale, and are shielded from the extreme weather variations outside the hive. In this work, honeybee activity is monitored using data from sensors inside the hive, along with monitoring the information related to the seasons, time of the day, external weather and the size of hive. WE-Bee's deep learning algorithm is based on Bidirectional LSTM encoder-decoder and attention mechanism. The results from field deployments demonstrate that the system is capable of cumulative weight estimation over multiple weeks. The results also show that the use of appropriate features and the diversity of training data is essential for robust performance of deep learning in the wild for this application.
UR - https://aaai.org/Conferences/AAAI-22/
UR - https://practical-dl.github.io/2022/
UR - https://practical-dl.github.io/2022/long_paper/18.pdf
UR - https://www.researchgate.net/publication/358954859_WE-Bee_Weight_Estimator_for_Beehives_Using_Deep_Learning
M3 - Conference paper
BT - AAAI Conference on Artificial Intelligence 2022
T2 - 36th AAAI Conference on Artificial Intelligence
Y2 - 22 February 2022 through 1 March 2022
ER -