General farm labor is in the $10.00 per hour range, and specialized labor, that is people trained in technical, biological or mechanical skills cost even more. Moreover, the cost of training increases also, since trainers are, in agricultural settings, generally part of the production crew taken out of production. Any crop production business that depends on honey bee pollination will be improved if the beekeeper who manages the bees used for pollination can save costs, or improve profit. The development of a handheld device that could measure a honey bee colony's condition without having to open a colony, without having to train the person what they were seeing when that colony was opened, and without the time and effort necessary to manipulate each frame in each colony would be a godsend to that industry. Examination time for entire beeyards by unskilled employees would be nearly the same as examination time for 1 or 2 pallets (4 or 8 colonies) now by trained employees. The savings would be incredible, the efficiency improved beyond imagination, and the costs of these tasks reduced to nearly zero. Moreover, the results of these examinations would be consistent, reliable and repeatable, three variables that are less certain when human skills enter the picture. OBJECTIVES: Bee Alert Technology, Inc. has an Army Phase II award that is investigating the use of sonograms of honey bee colony sounds for rapid detection of air toxicants. Bee colonies alter sound output within 30 seconds of an exposure event, and the resultant sonograms change in ways that are characteristic of the type of chemical involved (i.e., provide a sonic fingerprint). The USDA project focuses on research on natural pests and diseases of bee colonies. The research includes field trials with bee colonies, as well as bench top work on the most appropriate data analysis methods for rapid detection of bee pathogens. Phase I will culminate with a design for a hand-held sensor. Specific research objectives include: 1) Verification of sonic detection of varroa mites, africanization of colonies, and queenlessness, 2) Exploration of expansion of acoustic methods to the discovery of foulbrood, hive beetle, and possibly pesticide exposure, and 3) Development of algorithms using statistical or artificial neural network (ANN) methods for sonographic analysis. APPROACH: For varroa mites and American Foul Brood, we will work with scientists who maintain well-characterized colonies for experimental work. Primarily, this will include continued work with Jeff Harris and Stephen Pernal. They have both agreed to collect data or allow our technicians to collect data, at no cost. For sampling Africanized bees and hive beetle, we will work with volunteers, including other researchers and beekeepers. We already have a list of researchers and beekeepers in southern states who are more than willing to record sounds from Africanized hives (in Texas) and beetle infested hives, in Ohio, Georgia, and the Carolinas. For data collection for verification and expansion of acoustic profiling we have put together an acoustic sampling kit consisting of a digital Marantz PMD670 recorder, a microphone amplifier (SME 2100) from Saul Mineroff Electronics and two 1/8 inch diameter, 12 inch long, probe microphones. Each microphone is a high gain electret condenser microphone, one to be used for control hives, the other for mite infested or diseased hives. For meteorological data, we include a data-logging Kestrel 4000, weather meter. The kit is shipped in a water-tight, tough, polyethylene case, Pelican style. Guidelines for data collection are that the colonies need to be visually inspected and ranked according to degree of pathology (e.g., mite level, number of diseased cells, and scales, etc.) after sampling. For each hive, a two minute audio sample should be recorded in mid-morning, mid-afternoon, and after dark (when all of the bees are back in the hive for the night). Where possible, there should be a balanced data set of control and affected hives, or minimally three controls and ten stressed hives. Ideally, sampling would be conducted in the spring, mid-summer, and fall. When possible, we will gather data from colonies with the same type of problems, but different areas of the U.S. In all cases, the race of bees (if known) and basic meteorological data should be recorded. Filtered and summarized sonographic data are first analyzed using standard canonical discriminant function analysis to generate appropriate classification functions. The resulting unstandardized classification function coefficients are used as the basis for classifying subsequent sonographic data. Weighting new sonograms by the coefficient matrix produces a classification score which is then compared to a library of scores for known compounds to determine the most likely match. The exact method of interpreting sonograms is not restricted to discriminant analysis. For this USDA project, we will explore other techniques for sound pattern recognition that may be employed for this purpose. We anticipate that ANNs may prove to be useful, as well as additional statistical methods.
