Using YOLO in detecting and localising morepork sounds

Wang, Yanan (2021) Using YOLO in detecting and localising morepork sounds. Graduate student work (Unpublished)

PDF (Research report) - Accepted Version
[img] PDF (Deposit form) - Supplemental Material
Restricted to Repository staff only


Abstract or Summary

The research aims to implement and evaluate an advanced Convolutional Neural Network architecture, namely "You Only Look Once", to provide highly accurate data for the Morepork preservation projects. Implementing this technology is a complex task whose performance is affected by many factors. The research also seeks what the factors are and how they influence the results. This research initiates from exploring the Convolutional Neural Network's definition and its procedures for detecting bird sounds. After exploration, the Design Science Research methodology is determined to guide the research conduction. A Design Science Research frame is adapted with five stages: 1) Identify the problems and limit the research scope; 2) Conduct a systematic literature review and formulate a design for the problems; 3) Implement the artefacts according to the design; 4) Evaluate the artefacts' results; 5) Generate knowledge from the process. The literature review identifies three possible factors influencing the architecture's performance: presentation type, colourmap type, and CNN architecture. Accordingly, the author defines three research question with hypotheses to discuss the factors' influences. A full experimental design is conducted to verify the hypotheses. The research outcomes contain artefacts that contribute to the Morepork preservation projects and knowledge that directs future relevant research.

Item Type:Graduate student work
Keywords that describe the item:AI, CNN, colourmap, detection, image processing, Morepork, sound, spectrogram, YOLO, acoustic detection
Subjects:T Technology > T Technology (General)
Divisions:Schools > Centre for Business, Information Technology and Enterprise > School of Information Technology
ID Code:7805
Deposited By:
Deposited On:16 Aug 2021 00:27
Last Modified:16 Aug 2021 00:39

Repository Staff Only: item control page