Discipline: Computer Sciences and Information Management
Subcategory: Ecology
Darrion Long - Lincoln University, MO
Co-Author(s): Zachary Knuth, Derrick Parker, Jr., and David Heise, Lincoln University, MO Candace Galen and Johannes Schul, University of Missouri, MO Nicole Miller-Struttmann, Webster University, MO
Acoustic monitoring of pollinators is a topic of growing interest in field ecology and pollinator conservation. Given the importance of pollinators to agriculture, we believe development of acoustic monitoring methods may lead to significant benefits for farmers and others who rely on pollination services. This research focuses on audio detection and analysis based on acoustic monitoring techniques and is used to detect the sound produced by the oscillating movement of bee wings. Determining how to develop an algorithm that is as accurate as the human auditory system may prove useful in studying ecosystems. The currently developed algorithm is able to detect sounds from audio recordings of bees in their natural habitat. This algorithm could be later expanded to include other types of flying insects or wildlife that emit sounds during movement. The current work is to validate the algorithm by comparing the output of the algorithm to manual annotations of the data. Our trial conducted on previous data had many ’false positives’ which may indicate the algorithm’s ability to have a more sensitive perception of hearing than observed during human annotations of sounds recorded. If we consider the possible ways to classify the detection results of the algorithm, we can represent the results using a confusion matrix. The four categories are: 1) true positives, which indicate correct detection of buzzes compared to the ground truth, 2) false positives, which indicate detection of buzzes outside the ground truth, 3) true negatives, when neither human annotation or algorithm detects a buzz, and 4) false negatives, where the algorithm misses the detected buzzes in the ground truth. Currently, we are sorting through a previously annotated set of ground truth in an attempt to discover exactly what is happening at each of the ‘false positives’ and ‘false negatives’ and to verify their accuracy. When complete, it is our hope that this algorithm can be implemented as an aid in the conservation of native pollinators. In the future, we may be able to extend this method to gather statistical data about diversity of pollinators in a specific area without disrupting their natural habitat.
Funder Acknowledgement(s): This study was supported by the National Science Foundation: (HBCU-UP) Award # HRD-1410586 to David Heise, PhD, Lincoln University, Jefferson City MO; and Award # IIA-1355406 to John Walker, University of Missouri, Columbia, MO.
Faculty Advisor: David Heise, HeiseD@lincolnu.edu
Role: The duties given as a research assistant during the course of this research is primarily data analysis. The manual annotations are viewed for sound events outside the oscillating movement of bee wings, and anything before and after the observation period, which were omitted leaving the ground truth desired. After, the algorithm is run to see if it can detect results similar to the ground truth, then the dissimilar results are individually examined to determine if the algorithm performed better or worse than the original human annotation.