00000nam 2200000zi 4500 9.875215 CaOODSP 20221107164410 m o d f cr cn||||||||| 190620t20182018nsca ob f000 0 eng d CaOODSP eng rda CaOODSP eng eng fre n-cn--- D68-9/136-2018E-PDF Latta, Hope, author. Investigation to replace machine learning architecture in DRDC's aural classifier / Hope Latta and Carolyn M. Binder, DRDC - Atlantic Research Centre. Dartmouth, Nova Scotia : Defence Research and Development Canada = Recherche et développement pour la défense Canada, 2018. ©2018 1 online resource (vi, 20 pages, 2 unnumbered pages) : illustrations (chiefly colour). text txt rdacontent computer c rdamedia online resource cr rdacarrier Reference document ; DRDC-RDDC-2017-D136 Cover title. "Can unclassified." "February 2018." Includes bibliographical references (page 20). "Passive acoustic monitoring (PAM) is often used to detect the presence of whales. This is a meaningful way to learn about the size of their community, travel patterns, and reactions to human activity. Knowing when whales are nearby is also a significant part of avoiding collisions between whales and ships, and reducing the impact of sonar. However, PAM can be subject to large false detection rates. Machine learning, the algorithmic problem behind improving artificial intelligence, is able to increase the accuracy of PAM by automating the classification process. It is able to provide real-time, 24-hour monitoring, which would previously have been conducted by a human analyst. The purpose of this project was to replace the classifier architecture in DRDC's aural classifier with one that results in an increased performance. In order to choose the most suitable classification model type for a set of right, blue, fin, and humpback whale vocalizations, the K-Nearest Neighbours and Support Vector Machine models were evaluated. A comparative study of the literature associated with both classifiers provided benefits and drawbacks to examine before moving forward. The algorithms were trained using normalized perceptual signal features extracted by DRDC's aural classifier, then compared for evaluation. Preliminary results showed that the current implementation of the DRDC aural classifier produced the best performance"--Abstract, page i. Includes abstracts in English and French. Binder, Carolyn M., author. Defence R&D Canada. Atlantic Research Centre. Reference document (Defence R&D Canada) DRDC-RDDC-2017-D136. (CaOODSP)9.833971 PDF 1.18 MB https://publications.gc.ca/collections/collection_2019/rddc-drdc/D68-9-136-2018-eng.pdf