Graduate Thesis Or Dissertation
 

Multi-instance multi-label learning : algorithms and applications to bird bioacoustics

Public Deposited

Contenu téléchargeable

Télécharger le fichier PDF
https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/t148fm24z

Descriptions

Attribute NameValues
Creator
Abstract
  • We consider the problem of supervised classification of bird species from audio recordings in a real-world acoustic monitoring scenario (i.e. audio data is collected in the field with an omnidirectional microphone, without human supervision). Obtaining better data about bird activity can assist conservation efforts, and improve our understanding of their interactions with the environment and other organisms. However, traditional observation methods are labor- intensive. Most prior work on machine learning for bird song is not applicable to real-world acoustic monitoring, because it assumes recordings contain only a single species of bird, while recordings typically contain multiple simultaneously vocalizing birds. We propose to use the multi-instance multi-label (MIML) framework in machine learning for the species classification problem, where the dataset is viewed as a collection of bags of instances paired with sets of labels. Furthermore, we formalize MIML instance annotation, where the goal is to predict instance labels while learning only from bag label sets. We develop the first MIML representation for audio, and several new algorithms for MIML instance annotation based on support vector machines or classifier chains. The proposed methods classify either the set of species present in a recording, or individual calls, while learning only from recordings paired with a set of species. This form of training data requires less human effort to obtain than individually labeled calls. These methods are successfully applied to audio collected in the field which included multiple simultaneously vocalizing species. The proposed algorithms for MIML classification are general, and are also applied to object recognition in images.
License
Resource Type
Date Available
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Academic Affiliation
Non-Academic Affiliation
Subject
Déclaration de droits
Publisher
Peer Reviewed
Language
Replaces

Des relations

Parents:

This work has no parents.

Dans Collection:

Articles

La vignette Titre Date de téléchargement Visibilité actes
file details BriggsForrestS2013.pdf 2017-08-17 Public Télécharger