The goal is to inexpensively and automatically deduce bee behavior by analyzing the sounds made inside the hive.


Problem

It is not easy to understand what goes on inside a bee hive. Bee hives are too dark, cramped, and crowded to place cameras. Every motion of every bee – to some extent – results in a vibration in the air or on the surface of the comb. Is there anything about bee behavior that can be deduced from their induced acoustic vibrations? And if there is, can we do it on the cheap?

Speaking of cheap, it used to cost the military tens of millions to dollars for custom hardware and teams of programmers in order to calculate Fast Fourier Transforms for spectral analysis radar and sonar applications. Moore’s Law, Open Source software, and cheap ($99) digital field recorders provide superior and faster tools that anybody (with a little knowledge) can use to analyze colony sounds.


Prior Art

There is a surprising list of behaviors that are claimed to be deduced from the analysis of bee sounds. I don’t know yet if all of them are true but my goal is to find out. Results of acoustic analysis include:

  • detect every time a bee flys out of a hive (bee counter)
  • detect a sick hive and diagnose the disease
  • detect when there are two or more queens in a hive
  • detect when there is no queen laying eggs in the hive
    • she could be dead (rolled or dropped by a clumsy beekeeper or balled by her sisters)
    • she could be packing her bags (getting ready to swarm)
  • detect and identify a chemical spill or spraying or the release of an airborne Weapon of Mass Destruction

I will describe each of these scenarios first. Then I will describe how to implement a system that detects each event.

Apidictor

The Apidictor is an electronic device that indicates when the queen is not laying eggs. This is important because there is usually only one queen in a hive to lay eggs and worker bees only last 15-38 days in summer [SAMM98]. So a hive cannot last long without a queen. This usually happens when a colony is preparing to swarm but it can be caused by the death of a queen. A queen can be dead because

  • she’s old, she doesn’t smell like a Queen, her 50,000 daughters want her dead, and they want a younger sister to take her place
  • the beekeeper rolled (crushed) her
  • the beekeeper did not hold the frame she was on over the hive (dummy!) and she fell to the ground instead of the hive (and fertile queens cannot fly).

    apidictor

    apidictor

Whether it is a pending swarm or a dead queen, it is not good. And if it is a swarm, it would be really great to know ahead of time so you could prevent it. That is what the Apidector does.

The Apidictor was invented, patented, and marketed by the late Eddie F. Woods in the 1960s. He based it on the following insights into the lives of the bees:

  • As bees get older, they change jobs in the hive. Nurse bees are 4-1/2 to 6 days old. There are about 4000 nurse bees in a normal colony. Half feed the brood and half feed the queen who eats several times her weight in a single day.
  • When a colony prepares to swarm, the workers put the queen – who is too fat to fly – on a diet. This puts roughly 2000 nurse bees out of work and reduces the queen’s egg laying. Reduced egg laying results – in a couple days – in fewer larvae to feed which puts more nurse bees out of work. These girls don’t have lips so they stand around and flap their wings [Ok, I kinda made up that last part myself. Actually, they are fanning].
  • The wings of young bees do not harden until the 9th day – this makes them beat their wings faster than adult bees. Wings in flight beat faster than stationary bees anchored on comb fanning with their wings. Eddie made the following base frequency estimates:
    • 285 Hz – 4-1/2 day bee fanning
    • 250 Hz – Adult (9+ day) bee flying
    • 225 Hz – 6 day bee fanning
    • 190 Hz – Adult (9+ day) bee fanning
  • You can’t tell the adult bees to stop flying but if you measure collective frequency at night when no flying is taking place (or inside the hive), you can detect the extent to which nurse bees are out of work by analyzing the frequency band between 225 and 285 Hz.. This indicates trouble of one form or another.

He referred to this characteristic sound as a warble. But there was another sound he describes in the patent – a hiss:

If a healthy normal colony be disturbed by, say, a blow or tap, it will immediately respond with a short sharp hiss, which soon ceases. This reaction can be evoked as many times as desired. Its presence is an indication that there is in the colony a healthy queen. A sluggish rise in hiss, with a slow fade, indicates abnormality. It has been found that by detecting and differentiating between the warble and the hiss, an indication can be obtained which renders the opening of normal colonies unnecessary.

Eddie claimed that he could predict a swarming by three weeks and the building of queen cells by 10 days. I don’t know if any of this is true but it is all testable. And if it is found to be true, then it is a brilliant discovery. For more information on Eddie’s invention, visit Beesource Beekeeping at http://www.beesource.com/build-it-yourself/apidictor/. To view the Apidictor Patent click here.

Hello, I am the Queen and I will Kill all Rivals! 

The Feminine Monarchie, 1609

The Feminine Monarchie, 1609

If a hive is not queenright, which is beespeak for ‘queenless’, the worker bees grow several new queens. But if two or more queens hatch and encounter each other, they will fight to the death. This is not good because they could both die (bad for colony) and the loss of the second queen would remove any opportunity to swarm (how the super-organism reproduces). So evolution have given the queens two special sounds referred to as piping:

  • The newly emerged queen gives a tooting sound. This informs the rest of the girls that the queen has arrived. She makes this sound while roaming around the colony looking to kill any other emerging virgin queens. The Toot typically lasts 5 seconds with a single long syllable followed by several short syllables. The fundamental frequencies range from 350 – 550 Hz.
  • In response to the toot, another queen, still imprisoned in her plush queen cell, calls back with a quack. It has been suggested that this quack informs the colony to protect her so that she can emerge and perhaps abscond with a swarm of her own. The quack typically lasts more than 10 sec long as a series of short syllables. The fundamental frequencies range from 200 – 350 Hz.

Queens produce piping signals (roughly a G sharp) by pressing their thorax to the comb of the nest and rapidly contracting their thoracic flight muscles. Her wings are disengaged from the flight muscle while piping and so the wings move very little. By the way, this disengagement of wings from muscles is also used by heater bees to generate heat for the brood during cold nights [TAUTZ2008].

Piping is hardly a new discovery. Charles Butler wrote a treatise on bees entitled The Feminine Monarchie, shown at right (click to see a larger image) in the 1609 edition. He transcribed into musical notation the ‘piping’ and ‘quacking’ sounds produced by rival queens in a hive. He shows it in G rather than G sharp.

Now lets hear some of these quacks and toots. The recording below has several good quack from 0:00 to 0:40 and a good toot at 1:14.

Click here for piping sounds.

This recording is provided here by permission of Morris Ostrofsky. You can visit his piping web page.

Acoustic Sensor for Beehive Monitoring 

Spectrum of bee lift off

Spectrum of bee lift off

A patent by Trenton J. Brundage of Sherwood, Oregon has just been granted on March 11, 2010 (click here to read it) that claims an invention for assessing the productivity of a pollinating hive by counting bee sorties. It appears that when bees take off in flight, they ‘rev their engines’ so to speak. Once airborne, they ‘throttle back’ on the gas. In an airplane, a rev’ed engine has more revolutions of the propeller per minute. This causes the sound it makes to be higher in pitch than an engine that is ‘throttled back’. Likewise, when bees [I am assuming worker bees here since they are much more numerous than drones] take off, they beat their wings at 230 to 260 beats per second but after one tenth of a second, they slow their beats to a steady state of 180 to 210 beats per second.

The sonogram shown at right is a rather idealized visualization of this take off. The parallel lines are the harmonics [What are harmonics?]. The bottom line represents the fundamental frequency – the actual number of wing beats per minute. In reality, you have multiple bees taking off at the same time as well as bee flying around and birds and lots of other things. This is shown in the sonogram below. There are somethings we can do to get rid of bird sounds and other non-bee sounds. We solve that below below.

Honey bee acoustic recording and analysis system for monitoring hive health

I’ve saved the most intriguing technology for last. This patent claims that honey bees produce unique acoustic signatures when they are exposed to sub-lethal concentrations of airborne toxicants (chemical and biological) and stressors such as preditory mites. The key word here is unique. A signature for ammonia is different from toluene or airborne anthrax or sarin gas or varroa mites. If true, this gives us a way to detect and recognize a potentially dangerous event within a 3 mile radius.

If I lived within three miles of any factory or plant, I would want a bee hive monitoring air quality to notify me of a dangerous release. Of course, if you were living right next to the Union Carbine pesticide plant in Bhopal, India, your bees might not even make it back. The exposure has to be sub-lethal.

So this is how the patent claims that the invention works: put a microphone in a hive, take acoustic samples, generate a sonogram [What is a sonogram?]. From multiple sonograms, generate a baseline or normalized sonogram. Then expose the hive to various airborne toxicants or diseases and generate sonograms for them. You will end up with a library of sonograms for different scenarios. Now we apply statistical discrimination to determine what frequencies and amplitudes are most important for recognizing one of our scenarios. This is the first place where this patent leaves a lot of detail out. But they claim a 99.7 percent recognition rate based on only 3 discriminant functions! That is pretty amazing.

The big question I have that is not answered in the patent: what is the physical explanation for such a wide range of unique profiles? I have written the applicant but gotten no answer. In all the examples above, there was a clear, physical explanation of why a particular sound was made. But why does naptha produce a different acoustic signature from toluene?

Click here to review the patent.


Part 1 Summary

This finishes the present discussion of the biology behind this project. In the next section, I will outline how I propose to automate the recognition of these different acoustic events.

I welcome your comments and suggestions. Please post them to the Acoustic Analysis Project stub.

Continue to Part 2