Anomaly Detection in Beehives using Deep Recurrent Autoencoders

Abstract

Precision beekeeping allows to monitor bees' living conditions by equipping beehives with sensors. The data recorded by these hives can be analyzed by machine learning models to learn behavioral patterns of or search for unusual events in bee colonies. One typical target is the early detection of bee swarming as apiarists want to avoid this due to economical reasons. Advanced methods should be able to detect any other unusual or abnormal behavior arising from illness of bees or from technical reasons, e.g. sensor failure.In this position paper we present an autoencoder, a deep learning model, which detects any type of anomaly in data independent of its origin. Our model is able to reveal the same swarms as a simple rule-based swarm detection algorithm but is also triggered by any other anomaly. We evaluated our model on real world data sets that were collected on different hives and with different sensor setups.

Anomaly Detection in Beehives: An Algorithm Comparison

“…We used the HOBOS hives for training and validation purposes. That is, we trained on Bad Schwartau and Würzburg in independent settings using the reported and found [6] swarming events. Explicitly, we built two setups: one with training the models on the normal behavior of Bad Schwartau, using its anomalous behavior as a validation set for the parameter search, and one with the training step consisting of the normal behavior of Würzburg, while validating on its anomalous behavior set.…”

“…For the models Local Outlier Factor, Elliptic Envelope, Isolation Forest and One-Class SVM we relied on the implementations in [16]. In the same setting we searched for the remaining parameter α for the pre-trained AE, which we could not do in [6] due to missing labels. Within this (second) parameter setting step for the swarm detection, we used a windowing technique, shifting the window by 15 minutes forward in time to extract the next window.…”

“…We provide the recording time for all splits. [6] Predictions. RBA [24] is utilized on all anomalous behavior subsets to predict swarming events.…”

“…In this work we present an extension of our work in [6]. This includes a broader spectrum of anomaly detectors, not solely the recurrent autoencoder.…”

Detecção de Anomalias em Padrões Acústicos, de Temperatura e Umidade Sazonais para Abelhas Melíferas (Apis mellifera L.)

“…Desta forma, busca-se técnicas que façam uso intensivo dos dados provenientes da classe normal da colônia a fim de treinar modelos computacionais que reconheçam seu padrão de normalidade de modo que estejam prontos para sinalizar em suas saídas a presença de entradas distintas daquelas aprendidas. Esta classe de problemas é tratada por uma subárea de aprendizagem de máquina chamada de detecção de anomalias (anomaly detection) [Pimentel et al 2014] [Davidson et al 2020.…”