2016 12th International Conference on Intelligent Environments

Progress Towards an Intelligent Beehive

Building an Intelligent Environment to promote the Well-being of Honeybees

Donald Howard, Gordon Hunter, Olga Duran, Demetrios Venetsanos

Faculty of Science, Engineering & Computing,
Kingston University,
Kingston upon Thames, U.K.
[email protected] , [email protected] , [email protected] , [email protected]

Abstract— Honeybees are not only important for their conclude the paper with a summary of what has been achieved
production of honey but also, and more fundamentally, they are so far and suggesting future directions for the work.
almost certainly the most effective pollinators of crops. However,
there is strong evidence that the honeybee is under threat.
Various factors, including parasites, bacterial and fungal
II. HONEYBEE COLONIES AND PREVIOUS WORK MONITORING THEM
infections and pesticides have been suggested as causes of this. In
this project, we aim to promote bee well-being by encouraging A. Honeybees and Honeybee Colonies – key issues
bees to live more naturally, and propose an “intelligent beehive”, When it comes to pollination, it would be difficult to devise a
equipped with an array of appropriate sensors, to allow the more effective control system than that which already exists in
continuous monitoring of conditions in the hive in a non-invasive the functions of a honeybee colony. Properly understood, bees
manner. In this paper, we discuss the issues surrounding bee could also act as a “barometer” to environmental health. The
welfare and decline, present our innovative hive designs and
extraordinary cognitive abilities of honeybees have been
some preliminary results and analysis, concluding with future
work we plan to carry out, collecting & analysing data from a
successfully dominating pollination services through millions
range of hives equipped with a much broader range of sensors, of years of change, so it is a natural choice to let them manage
including a discussion of our equipment and methodologies. their own natural lifecycle and lifestyle as they see fit.

Keywords—honeybees; well-being; monitoring; intelligent The cognitive abilities of honeybee colonies are most fully
sensor network; beehive expounded by the works of Thomas Seeley, especially in his
books “Honeybee Ecology” [1] and “The Wisdom of the
I. INTRODUCTION Hive” [2]. In short, the colony is constantly integrating
The honeybee (Apis Mellifera) is vital to ecology and information from the environment to make decisions about
agriculture through its phenomenal ability to pollinate crops maximising growth, such as maintaining the hive at a constant
and other plants. However, as has recently received widespread temperature of 34°C during hot days and cold nights while the
publicity in the media, the honeybee has been in serious decline brood is developing, to the amount of brood being laid against
in many countries over recent decades, which has been the value of resources available to the colony. All this is being
attributed to a wide variety of factors, including the use of done against the constant changes going on in the environment
pesticides and parasites such as the Varroa mite. Nevertheless, as different flowers come in and out of bloom. Honeybees
the practices of conventional beekeeping, involving frequent respond quickly and efficiently to these changes, and their
invasive inspections of hives, interfering with and stressing the methods for doing so are well understood.
bees, and potentially spreading infections and parasites, quite
possibly contribute to this decline. In this paper, we propose an
alternative approach to beekeeping – using an intelligent B. Previous Work on Monitoring Honeybee Colonies
monitoring system within the hive to minimize the frequency Aristotle (c 384 – 322 BC) is credited as first noting that
of inspections and encourage the bees to live as naturally and honeybees perform a bizarre song and dance and speculating
with little interference from humans as possible. that it was some form of communication. The first full length
The remainder of this paper is structured as follows. The book on beekeeping in English was Charles Butler’s 1623
next section outlines some of the key issues surrounding work "Feminine Monarchy". However, it was the famous
honeybee colonies and beekeeping, the modern problems German ethologist Karl Ritter von Frisch (1886 – 1982) who
associated with these, and previous work on monitoring first unravelled the hidden messages encoded in the
various aspects of the environment of a beehive. We then mysterious “waggle dance” performance [3,4]. The Nobel
proceed to describe our novel hive design, and propose two Prize for Physiology or Medicine 1973 was awarded jointly to
possible approaches to monitoring a bee colony inside it. A Karl von Frisch, Konrad Lorenz and Nikolaas Tinbergen "For
number of our hives have recently (March 2016) been their discoveries concerning organization and elicitation of
deployed at various locations in the local area, and data from individual and social behaviour patterns (of animals)" (Nobel
these is already being collected. We subsequently propose Prize citation, 1973). Frisch’ s work was continued by his
methodologies for analyzing and interpreting the data, and student Martin Lindauer, who in turn was a mentor to Thomas

978-1-5090-4056-8/16 $31.00 © 2016 IEEE 262

DOI 10.1109/IE.2016.60
Seeley, who has written or co-written over 150 scientific III. NOVEL BEEHIVE DESIGN AND MONITORING SYSTEM
publications on honeybees, and to this day continues In this section, we describe our approach to addressing the
producing ground-breaking work on bees and their individual problems. In the next sub-section, we describe our novel hive
and social behavior, including “The Wisdom of the Hive” [2] design, whilst in the subsequent sub-section, we describe the
and “Honeybee democracy” [5]. In “Honeybee democracy”, sensor system included for monitoring the hive, its bee colony
he makes an exciting comparison of the swarm as a cognitive and the surrounding environment. Twelve of our hives have
entity, identifying the honeybee to a neural component of been deployed around the local area, at nine different locations
swarm intelligence [5]. all within a 10 mile (16 km) radius of the main Kingston
University campus. This area of deployment allows use of a
In 1997, Halit et al produced a paper [6] in which attempts are variety of suburban and rural situations, whilst keeping
made to control queen bee behavior with artificial sounds. environmental factors such as ambient temperature and
This is an interesting departure from previous work which rainfall reasonably uniform. A further four hives have been
seemed to focus on simply understanding the bee language. deployed at other locations, further away, but still in Southern
England, at sites in Surrey, Sussex and Devon.
The first person to develop an electronic device for listening to
bee sounds was Edward Farrington Woods (1901-1976), who A. Novel Hive Design
produced the “apiductor” – a sound analysis system As noted above, traditional beehives are designed so as to be
specifically designed for bee acoustics [7]. easy for beekeepers to make inspections and remove
honeycombs. However, our primary aim is to promote bee
The first scientific study to address the actual monitoring of well-being rather than make the work of beekeepers easy. We
bees appears to be by Ferrari et al in 2008 [8] and is have followed some of the observations made by Seeley [2]
significant to this report. The principal method used in their concerning what types of “nest sites” or hives tend to be
work was the use of Power Spectral Density (PSD) and preferred by honeybee swarms seeking a new home – namely
spectrograms, although the temperature was monitored as that the hive should have a volume of about 40 litres, and that
well. The results of the investigation revealed significant the entrance to the hive should have an area of around 12.5
acoustic characteristics in the immanent lead up (1/2 hour cm2 and be South-facing. However, we have compromised on
prior) to a swarming event. They had in fact identified the some of Seeley’s principles – notably that the hive should be
buzz running signal, so the signals identified could not be used around 5 metres off the ground. Whilst this may be
as a long term prediction of swarming advantageous in deterring predators, honey thieves, etc., it is
impractical from the point of view of making the hive
Work by Bencsik et al [9] again used spectrographic analysis accessible for maintenance and inspection. Our design instead
to study the acoustic data, making use of Principal hosts the hive around 1.5 metres off the ground – much less
Components Analysis (PCA) and eigenspectra to identify and than proposed by Seeley, but considerably more than the 30 –
classify the sounds. One interesting and unusual aspect of 50 cm commonly used by beekeepers, and may be sufficient to
their work was the use of accelerometers in place of deter ground-based predators.
microphones to gather their acoustic/vibration data.

Other work has been performed on species different to

honeybees. Mhatre et al [10] applied similar techniques to
sound reception and perception by small insects, as do
Reynolds & Riley [11]. Boucher et al studied the interactions
between worker and drone honeybees [12], while Pace et al
[13] used Hidden Markov Models to analyse the acoustics of
the sounds used by humpback whales to communicate
between each other.

Our initial work in this area [14, 15] focused on using acoustic
signals produced by a honeybee colony in an attempt to
identify whether the colony did or did not include a healthy Fig. 1 Seeley’s findings on honeybee nest site preferences (from [2]).
queen bee – the former being called “queenright” and the
latter “queenless” states. In those studies, we used spectral
analysis and a Self-Organising Map [16] to classify and
distinguish between the two states.

263
 and the likelihood for the bees to swarm; calibration of a gas
detector in order to detect and monitor bee pheromones could
be useful, but it is not obvious to what extend this would be
possible; finally, video monitoring could yield very useful
information, but would require very large amounts of data
storage, so is not considered a top priority for the time being.

The hives fitted with the Arnia sensors were deployed in mid-
March 2016, in good time to anticipate the 2016 swarming
season (in the mid-late Spring). However, very little data has
been acquired at the time of writing, and nothing warranting
detailed analysis so far, but data is now coming in (June 2016)
and it is hoped we will have enough to perform meaningful
investigation and pattern recognition soon.

(a) (b) (c)

(d)
Fig. 2 Design of the physical body of our novel hive : (a) Indivdual
components : cover, body and stand; (b) Assembled, but with front cover
removed for inspection; (c) Assembled, with full cover fitted; (d) Design of
components all on a single sheet of weather-resistant fibreboard, ready for
laser cutting.
Fig. 3. View of GUI display for monitoring four of our hives deployed in the
B. Sensors for Hive Monitoring field, using Arnia sensor technology. The numbers in the yellow hexagon
indicates the temperature in the hive’s “brood box”, and in the red
In the case of the hives already deployed, we are employing a “thermometer” the ambient temperature in the hive. The number in the blue
network of sensors designed and implemented by the drop shape is the relative humidity (as a percentage) and the green “towers”
company, Arnia [17], who have been designing and indicate the levels of sound. The black “weight” symbol indicates the current
implementing hive monitoring systems for several years. mass or weight of the hive realtive to when it is empty (only the rightmost
hive contained any bees at that time). The “bee” pattern indicates the level of
However, Arnia’s systems and services come at a substantial hive activity, and other symbols show the recent and current weather
cost, which may be prohibitively high for many beekeepers. conditions at that site, the wireless connection signal strength and the system’s
Thus, we have also designed our own bespoke sensor network, battery level.
and we intend to test and evaluate both systems, comparing
them for cost, effectiveness (in terms of efficient monitoring IV. PROPOSED METHODOLOGIES FOR ANALYSIS
of the hives) and value for money. The sensors in our bespoke Although it is difficult to be precise about the
system will monitor temperatures, mass, sound, relative methodologies which will be most appropriate for the analysis
humidity, gas (pheromone) level, air pressure, and video of the data until a substantial quantity of it has been acquired,
images. We believe that, in terms of utility for our aforestated in this section we describe methods likely to be suitable for
aims, the most important sensors, in decreasing order of analyzing the data expected to be obtained in this study.
importance, are weather sensors (external temperature, rain The data will be intrinsically somewhat heterogeneous –
and wind monitoring), which monitor conditions influencing including temperatures, mass, sound, relative humidity, gas
availability of forage, ease of flying, etc.; hive mass (pheromone) level, air pressure, and video images. However,
monitoring (which can be used to keep track of honey and the data are in the form of times series, even if different
brood production, and help predict when a swarm night quantities are sampled at different frequencies. These different
occur); internal temperature and humidity can be very sampling rates will mean that comparison between the various
important for making inferences about the activity of the series will have to be carried out in terms of the real time which
colony and the likely well-being of the brood; and acoustic has elapsed since some reference point, rather than simply the
sensors (microphones) can be used to detect queenlessness, number of samples. It is expected that each of the signals will

264
include periodic components – for example, there will almost in our study, and Arnia Ltd. for co-operation and providing
certainly be a daily component (with period 24 hours) to the some equipment at a reduced cost.
ambient temperature – so methods such as autocorrelation,
Fourier and cepstral analysis will be appropriate for identifying REFERENCES
these. Furthermore, it is expected that there will be relationship [1] Seeley, T.D. (1985) “Honeybee Ecology”
between the different series – for example, there may be a [2] Seeley, T.D. (1995) The Wisdom of the Hive. Cambridge, Massachusetts,
relation between the temperature and the sound level in the USA: Harvard University Press.
hive. However, these signals may not rise and fall at exactly the [3] Frisch, K. (1955) The Dancing Bees (Translation from original German
same times – there may be a time lag between a change in one edition)
and the corresponding change in the other. This will require [4] Frisch, K. (1993) The Dance Language and Orientation of Bees
(Translation from original German edition)
cross-correlation analysis, allowing for the inclusion of a time
lag between the signals. [5] Seeley, T.D. (2010) Honeybee Democracy. Princeton, USA: Princeton
University Press.
Once one aim of this project is to predict notable events in [6] Halit, E., Whiffler, L. & Maning, R. (1997) Electronic sensing and
the bee colony’s years, such as when the colony is about to identification of queen bees in honeybee colonies. In IEEE Conference om
swarm, intelligent pattern recognition algorithms will be Instrumentation and Measurement Technology Conference. Ottawa, Canada.
required. In our previous paper on this topic [15], we employed [7] Boys, R., 1999. Listen to the Bees. [Online] Available at:
http://www.beedata.com/data2/listen/listenbees.htm (Accessed March 2016).
a self-organising map [16] to explore patterns in the sounds
[8] Ferrari, S., Silva, M., Guarino, M. & Berckmans, D. (2008) Monitoring of
produced by the colony over time. However, this approach had swarming sounds in bee hives for early detection of the swarming period.
various limitations, and a wider range of machine learning Computers and Electronics in Agriculture, Vol. 64, pp.72-77.
pattern recognition methods [18] should be explored. [9] Bencsik, M. et al., (2011) Identification of the honey bee swarming
Analogies could be made with the types of approach used to process by analysing the time course of hive vibrations. Computers and
identify temporal patterns associated with the “Activities of Electronics in Agriculture, Vol. 76, pp.44-50.
Daily Living” (ADL) of people [19] – for example, Hidden [10] Mhatre, N., Montealgre-Z, F., Balakrishnan, R. & Robert, D. (2012).
Markov Models [20, 21] have been extensively used to Sound reception and radiation in a small insect. In Proceedings of Acoustics
recognize patterns, both in human speech and in other “state 2012 : Joint Counference of the Societé Française d’Acoustique and Institute
of Acoustics, Nantes, France, April 2012.
based” systems (e.g. [22]).
[11] Reynolds, D.R. & Riley, J.R., 2002. Remote-sensing, telemetric and
V. SUMMARY AND FUTURE WORK computer-based technologies for investigating insect movement: a survey of
existing and potential techniques. Computers and Electronics in Agriculture,
The aim of this project is to present and implement an Vol. 35, pp.271-307.
original and valid solution to a real world problem, namely [12] Boucher, M. & Schneider, S.S. (2009) Communication signals used in
honeybee colony decline, which is being taken seriously by worker-drone interactions in the honeybee, Apis mellifera. Animal Behaviour,
government bodies in many countries. We have designed, built Vol. 78, pp.247-54.
and deployed several “smart hive”, with both a novel physical [13] Pace, F., White, P.R. & Adam, O. (2012) Hidden Markov Modeling for
humpback whale (Megaptera novaeangliae) call classification. In Proceedings
design and an intelligent sensor network, to allow non-invasive of Acoustics 2012 : Joint Counference of the Societé Française d’Acoustique
continuous monitoring of honeybee colonies to try to ensure and Institute of Acoustics, Nantes, France, April 2012.
their well-being and provide early warning of events such as [14] Howard, D. (2012) Audio techniques to determine the presence of the
swarming or problems such as infestation by parasites. Queen bee in a hive. Unpublished MEng Dissertation, Kingston University,
U.K.
However, at the time of writing, these “smart hives” had
only recently been deployed in the field, so to date very little [15] D. Howard,O. Duran, G. Hunter & K. Stebel (2013) “Signal Processing
the Acoustics of Honeybees (Apis Mellfera) to Identify the ‘Queenless’ State
data has been acquired. The analysis and interpretation of the in Hives”, Proceedings of the Institute of Acoustics, Vol. 35. (1) pp 290 – 297
data therefore remains work for the future. We also recognise [16] T. Kohonen (1995) “Self-Organizing Maps”, Springer Series in
that the philosophy of our proposed approach – to promote a Information Sciences, Berlin, Heidelberg. (Third Extended Edition, 2001)
more natural lifestyle for honeybees, focusing on their role as [17] Arnia Ltd. (2016) http://www.arnia.co.uk/ (Accessed April 2016)
pollinators rather than on honey production – may meet some [18] T. Mitchell (1997) “Machine Learning”, McGraw Hill, New York
resistance from established conventional beekeepers. [19] K. Avgerinakis, A. Briassouli & K. Kompatsiaris (2015) “Activities of
Nevertheless, there may be room for compromise using a daily living recognition using optimal trajectories from motion boundaries”,
hybrid approach – perhaps making use of innovations such as Journal of Ambient Intelligence and Smart Environments, Vol. 7 (6), pp. 817-
the “flow hive” technology [23] recently developed by 834
beekeepers in New Zealand, which allows harvesting of honey [20] L. R. Rabiner (1989) “A Tutorial on Hidden Markov Models and and
from hives in a controlled and non-invasive way, causing much selected applications in speech recognition”, Proceedings of the IEEE, 77 (2),
pp 257 - 286
less stress and disruption to the bee colony than do
[21] J. Holmes & W. Holmes (2001) Speech Synthesis and Recognition,
conventional honey extraction methods. Taylor & Francis, London, U.K.
ACKNOWLEDGMENTS [22] G. Hunter, K. Zienowicz & A. Shihab (2008) “The Use of Mel Cepstral
Coefficients and Markov Models for the Automatic Identification,
Donald Howard is grateful to the SEC Faculty Doctoral Classification and Sequence Modelling of Salient Sound Events Occurring
School at Kingston for the award of a PhD Studentship. We During Tennis Matches”. The Journal of the Acoustical Society of America
would also like to thank Sivakhani Sivanesan and the (JASA), Vol. 123(5), p. 3431.
Biodiversity and Landscape team at Kingston University both [23] Flow Hive promotional video
https://www.youtube.com/watch?v=WbMV9qYIXqM
for funding some equipment and allowing us to use their hives

265