Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Passive acoustic monitoring of animal populations with compressed sensing
Thesis (MSc)--Stellenbosch University, 2025.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Stellenbosch : Stellenbosch University
2026
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867614018279047168 |
|---|---|
| access_status_str | Open Access |
| author | Rasolofohery, Milanto Ferdinand |
| author2 | Dufourq, Emmanuel |
| author_browse | Dufourq, Emmanuel Rasolofohery, Milanto Ferdinand |
| author_facet | Dufourq, Emmanuel Rasolofohery, Milanto Ferdinand |
| author_sort | Rasolofohery, Milanto Ferdinand |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (MSc)--Stellenbosch University, 2025. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/134790 |
| institution | Stellenbosch University (South Africa) |
| last_indexed | 2026-06-10T12:45:21.489Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2026 |
| publishDateRange | 2026 |
| publishDateSort | 2026 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/134790 Passive acoustic monitoring of animal populations with compressed sensing Rasolofohery, Milanto Ferdinand Dufourq, Emmanuel Stellenbosch University. Faculty of Science. Dept. of Applied Mathematics. Human-animal communication Wildlife monitoring Acoustic emission testing Bioacoustics Thesis (MSc)--Stellenbosch University, 2025. Rasolofohery, M. F. 2025. Passive Acoustic Monitoring of Animal Populations with Compressed Sensing. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/b50e0ec5-46a5-4211-85a9-e387e9125953 ENGLISH ABSTRACT: Passive acoustic monitoring (PAM) is widely used in biodiversity research to study animal vocalizations non-invasively. However, modern PAM deployments often involve multiple recorders operating continuously over extended periods, generating large volumes of audio data that typically require posthoc analysis. This imposes significant challenges for real-time conservation decision-making and highlights critical limitations related to data storage, energy consumption, and hardware scalability. To address these challenges, this thesis investigated the use of compressed sensing (CS) as a low-power audio compression technique within PAM workflows and evaluated its impact on deep learning-based species detection. Experiments were conducted using two bioacoustic datasets containing vocalizations of the Thyolo alethe (Chamaetylas choloensis) and the Hainan gibbon (Nomascus hainanus), the world’s rarest primate. CS was applied to both raw audio signals and spectrogram representations, and its performance was compared to standard codecs, including MPEG-1 Audio Layer III, Free Lossless Codec, Advanced Audio Coding, Ogg, and spectrogram-based compression using autoencoders and Joint Photographic Experts Group. Convolutional neural networks (CNNs) trained on CS-compressed and reconstructed spectrograms achieved F1-scores within 1–5% of those obtained from uncompressed data, even at compression ratios as high as 80–90%. Furthermore, CS demonstrated F1-scores comparable to those of standard codecs while providing superior energy efficiency on low-power devices such as the Raspberry Pi, reducing both processing time and estimated carbon emissions. This study provides the first comprehensive comparison of CS-based and traditional audio compression methods in a bioacoustic detection context. The results suggest that CS offers a viable path toward real-time, scalable, and energy-efficient wildlife monitoring systems, with implications for enhancing conservation practices in remote and resource-constrained environments. AFRIKAANSE OPSOMMING: Passive acoustic monitoring (PAM) word grootliks gebruik in biodiversiteitsnavorsing om die vokaliesasies van diere op n nie-indringende manier te bestudeer. Die toepassing van moderne PAM behels dikwels die gebruik van verskeie klank opnemers wat oor lang tydperke aaneenlopend klank opneem. Hierdie opnames behels die versameling van groot hoeveelhede oudio wat gewoonlik na die tyd eers verder verwerk moet word. Dit stel egter beduidende uitdagings rakende die intydse besluitneming wat moet plaas vind in natuurbewaring. Verder beklemtoon dit ook die kritieke beperkings wat verband hou met die opberging van data, energieverbruik en hardeware-skaalbaarheid. Om hierdie uitdagings aan te spreek, het hierdie tesis compressed sensing (CS) as ’n lae-krag oudiokompressietegniek binne ’n PAM-werkvloei ondersoek. Die impak daarvan was geëvalueer op diep leer-gebaseerde spesie-opsporing. Eksperimente was uitgevoer met behulp van twee bioakoestiese datastelle wat vokalisasies van die Thyolo alethe (Chamaetylas choloensis) en die Hainan gibbon (Nomascus hainanus), die wêreld se skaarsste primaat, bevat. CS was toegepas op beide die onverwerkte oudioseine en spektrogramvoorstellings. Die behaalde prestasie daarvan was vergelyk met standaardkodeks metodes, insluitende MPEG- 1 Audio Layer III, Free Lossless Codec, Advanced Audio Coding, Ogg, en spektrogram-gebaseerde kompressie met behulp van autoencoders en die Joint Photographic Experts Group. Convolutional neural networks (CNNs) wat opgelei is op CS gekompresseerde oudio en gerekonstrueerde spektrogramme het F1-tellings binne 1-5% van dié verkry van ongekompresseerde oudio data behaal, selfs teen kompressieverhoudings so hoog soos 80-90%. Verder het CS F1-tellings behaal wat vergelykbaar is met dié van standaardkodeks, terwyl dit beter energie-doeltreffendheid op lae-kragtoestelle soos die Raspberry Pi bied, wat beide verwerkingstyd en geraamde koolstofvrystellings verminder. Hierdie studie bied die eerste omvattende vergelyking van CS-gebaseerde en tradisionele oudiokompressiemetodes in ’n bioakoestiese opsporingskonteks. Die resultate dui daarop dat CS ’n lewensvatbare pad bied na intydse, skaalbare en energie-doeltreffende wildlewe-moniteringstelsels, met implikasies vir die verbetering van bewaringspraktyke in afgeleë en hulpbronbeperkte omgewings. Masters 2026-01-08T06:26:08Z 2026-01-08T06:26:08Z 2025-12 Thesis https://scholar.sun.ac.za/handle/10019.1/134790 Stellenbosch University xiv, 171 pages application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Human-animal communication Wildlife monitoring Acoustic emission testing Bioacoustics Rasolofohery, Milanto Ferdinand Passive acoustic monitoring of animal populations with compressed sensing |
| title | Passive acoustic monitoring of animal populations with compressed sensing |
| title_full | Passive acoustic monitoring of animal populations with compressed sensing |
| title_fullStr | Passive acoustic monitoring of animal populations with compressed sensing |
| title_full_unstemmed | Passive acoustic monitoring of animal populations with compressed sensing |
| title_short | Passive acoustic monitoring of animal populations with compressed sensing |
| title_sort | passive acoustic monitoring of animal populations with compressed sensing |
| topic | Human-animal communication Wildlife monitoring Acoustic emission testing Bioacoustics |
| url | https://scholar.sun.ac.za/handle/10019.1/134790 |
| work_keys_str_mv | AT rasolofoherymilantoferdinand passiveacousticmonitoringofanimalpopulationswithcompressedsensing |