{"id":{"repo_id":"exeter","oai_identifier":"oai:figshare.com:article/32627343"},"canonical_url":"https://search.dev.ndltd.org/etd/exeter/oai:figshare.com:article/32627343","repository":{"repo_id":"exeter","name":"University of Exeter","base_url":"https://api.figshare.com/v2/oai"},"display":{"title":"Automating Video Analysis for Marine and Behavioural Research: Deep Learning Solutions to Analytical Bottlenecks","abstract":"Technological advances have dramatically increased marine data collection capacity, shifting the primary challenge in ocean science from data acquisition to data processing. An 'analytical bottleneck' now limits the extraction of ecological insight from information-rich datasets such as video recordings. This thesis develops and validates deep learning methods to address this bottleneck, demonstrating how YOLO-based object detection can accelerate inference in video analysis and enable novel analyses. Chapter 2 presents an automated pipeline for estimating abundance in Baited Remote Underwater Video (BRUV) data from offshore wind farms in the northern North Sea. BRUV surveys are widely used for marine biodiversity assessment but remain constrained by labour-intensive manual analysis. The YOLO-based detection model developed here automatically enumerates two commercially important taxa – Gadidae “cods” (mAP = 0.896 ± 0.009) and Pleuronectiformes “flatfish” (mAP = 0.814 ± 0.009) – replacing human analysts in MaxN estimation (the maximum individuals observed in a single frame). The automated pipeline also enabled investigation of stereo-vision revealing that combining viewpoints from stereo-BRUVs increases sensitivity to experimental variables. Chapter 3 presents AnimalTrackR (github.com/mariolambrette/AnimalTrackR), an open-source R package that enables researchers with limited programming experience to train custom YOLO detection models for laboratory behavioural studies. The package provides an integrated workflow from image annotation through model training to behavioural classification, removing technical barriers that limit adoption of deep learning methods. Functionality is demonstrated through three case studies tracking zebrafish (Danio rerio), rainbow trout (Oncorhynchus mykiss), and gilthead seabream (Sparus aurata). The zebrafish study validates the automated behavioural classification pipeline against manual observations. These chapters demonstrate that automated video analysis can reduce processing time while extracting higher-resolution data than manual methods permit. By providing accessible, well-documented implementations, this thesis aims to accelerate deep learning adoption in marine ecology and contribute to maximising the scientific value extracted from video-based research.<p></p>","abstract_html":"Technological advances have dramatically increased marine data collection capacity, shifting the primary challenge in ocean science from data acquisition to data processing. An &#x27;analytical bottleneck&#x27; now limits the extraction of ecological insight from information-rich datasets such as video recordings. This thesis develops and validates deep learning methods to address this bottleneck, demonstrating how YOLO-based object detection can accelerate inference in video analysis and enable novel analyses. Chapter 2 presents an automated pipeline for estimating abundance in Baited Remote Underwater Video (BRUV) data from offshore wind farms in the northern North Sea. BRUV surveys are widely used for marine biodiversity assessment but remain constrained by labour-intensive manual analysis. The YOLO-based detection model developed here automatically enumerates two commercially important taxa – Gadidae “cods” (mAP = 0.896 ± 0.009) and Pleuronectiformes “flatfish” (mAP = 0.814 ± 0.009) – replacing human analysts in MaxN estimation (the maximum individuals observed in a single frame). The automated pipeline also enabled investigation of stereo-vision revealing that combining viewpoints from stereo-BRUVs increases sensitivity to experimental variables. Chapter 3 presents AnimalTrackR (github.com/mariolambrette/AnimalTrackR), an open-source R package that enables researchers with limited programming experience to train custom YOLO detection models for laboratory behavioural studies. The package provides an integrated workflow from image annotation through model training to behavioural classification, removing technical barriers that limit adoption of deep learning methods. Functionality is demonstrated through three case studies tracking zebrafish (Danio rerio), rainbow trout (Oncorhynchus mykiss), and gilthead seabream (Sparus aurata). The zebrafish study validates the automated behavioural classification pipeline against manual observations. These chapters demonstrate that automated video analysis can reduce processing time while extracting higher-resolution data than manual methods permit. By providing accessible, well-documented implementations, this thesis aims to accelerate deep learning adoption in marine ecology and contribute to maximising the scientific value extracted from video-based research.&lt;p&gt;&lt;/p&gt;","abstract_has_math":false,"creators":["Mario Lambrette (21044177)"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026-06-09T00:00:00Z","date_published":"2026-06-09T00:00:00Z","updated_at":"2026-07-27T19:32:41Z","subjects":["YOLO","BRUV","Deep learning","Behavioural analysis","Zebrafish","Artificial Intelligence","Fish detection","Object detection","Computer Vision","Underwater Video","Animal Tracking"],"languages":[],"rights":["All rights reserved","Open Access after 2026-12-15"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["10779/exe.32627343.v1"],"render_values":[{"text":"10779/exe.32627343.v1","href":null,"code":true}]}]},"links":{"outbound_url":null,"outbound_label":null,"outbound_source":null},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:creator","label":"Author","values":["Mario Lambrette (21044177)"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2026-06-09T00:00:00Z"]},{"key":"dc:relation","label":"Dc Relation","values":["https://figshare.com/articles/thesis/Automating_Video_Analysis_for_Marine_and_Behavioural_Research_Deep_Learning_Solutions_to_Analytical_Bottlenecks/32627343"]},{"key":"dc:type","label":"Dc Type","values":["Text","Thesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["YOLO","BRUV","Deep learning","Behavioural analysis","Zebrafish","Artificial Intelligence","Fish detection","Object detection","Computer Vision","Underwater Video","Animal Tracking"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:rights","label":"Dc Rights","values":["All rights reserved","Open Access after 2026-12-15"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["10779/exe.32627343.v1"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Technological advances have dramatically increased marine data collection capacity, shifting the primary challenge in ocean science from data acquisition to data processing. An 'analytical bottleneck' now limits the extraction of ecological insight from information-rich datasets such as video recordings. This thesis develops and validates deep learning methods to address this bottleneck, demonstrating how YOLO-based object detection can accelerate inference in video analysis and enable novel analyses. Chapter 2 presents an automated pipeline for estimating abundance in Baited Remote Underwater Video (BRUV) data from offshore wind farms in the northern North Sea. BRUV surveys are widely used for marine biodiversity assessment but remain constrained by labour-intensive manual analysis. The YOLO-based detection model developed here automatically enumerates two commercially important taxa – Gadidae “cods” (mAP = 0.896 ± 0.009) and Pleuronectiformes “flatfish” (mAP = 0.814 ± 0.009) – replacing human analysts in MaxN estimation (the maximum individuals observed in a single frame). The automated pipeline also enabled investigation of stereo-vision revealing that combining viewpoints from stereo-BRUVs increases sensitivity to experimental variables. Chapter 3 presents AnimalTrackR (github.com/mariolambrette/AnimalTrackR), an open-source R package that enables researchers with limited programming experience to train custom YOLO detection models for laboratory behavioural studies. The package provides an integrated workflow from image annotation through model training to behavioural classification, removing technical barriers that limit adoption of deep learning methods. Functionality is demonstrated through three case studies tracking zebrafish (Danio rerio), rainbow trout (Oncorhynchus mykiss), and gilthead seabream (Sparus aurata). The zebrafish study validates the automated behavioural classification pipeline against manual observations. These chapters demonstrate that automated video analysis can reduce processing time while extracting higher-resolution data than manual methods permit. By providing accessible, well-documented implementations, this thesis aims to accelerate deep learning adoption in marine ecology and contribute to maximising the scientific value extracted from video-based research.<p></p>"]},{"key":"dc:title","label":"Title","values":["Automating Video Analysis for Marine and Behavioural Research: Deep Learning Solutions to Analytical Bottlenecks"]}]}],"canonical_facts":{"dc:creator":["Mario Lambrette (21044177)"],"dc:date":["2026-06-09T00:00:00Z"],"dc:description":["Technological advances have dramatically increased marine data collection capacity, shifting the primary challenge in ocean science from data acquisition to data processing. An 'analytical bottleneck' now limits the extraction of ecological insight from information-rich datasets such as video recordings. This thesis develops and validates deep learning methods to address this bottleneck, demonstrating how YOLO-based object detection can accelerate inference in video analysis and enable novel analyses. Chapter 2 presents an automated pipeline for estimating abundance in Baited Remote Underwater Video (BRUV) data from offshore wind farms in the northern North Sea. BRUV surveys are widely used for marine biodiversity assessment but remain constrained by labour-intensive manual analysis. The YOLO-based detection model developed here automatically enumerates two commercially important taxa – Gadidae “cods” (mAP = 0.896 ± 0.009) and Pleuronectiformes “flatfish” (mAP = 0.814 ± 0.009) – replacing human analysts in MaxN estimation (the maximum individuals observed in a single frame). The automated pipeline also enabled investigation of stereo-vision revealing that combining viewpoints from stereo-BRUVs increases sensitivity to experimental variables. Chapter 3 presents AnimalTrackR (github.com/mariolambrette/AnimalTrackR), an open-source R package that enables researchers with limited programming experience to train custom YOLO detection models for laboratory behavioural studies. The package provides an integrated workflow from image annotation through model training to behavioural classification, removing technical barriers that limit adoption of deep learning methods. Functionality is demonstrated through three case studies tracking zebrafish (Danio rerio), rainbow trout (Oncorhynchus mykiss), and gilthead seabream (Sparus aurata). The zebrafish study validates the automated behavioural classification pipeline against manual observations. These chapters demonstrate that automated video analysis can reduce processing time while extracting higher-resolution data than manual methods permit. By providing accessible, well-documented implementations, this thesis aims to accelerate deep learning adoption in marine ecology and contribute to maximising the scientific value extracted from video-based research.<p></p>"],"dc:identifier":["10779/exe.32627343.v1"],"dc:relation":["https://figshare.com/articles/thesis/Automating_Video_Analysis_for_Marine_and_Behavioural_Research_Deep_Learning_Solutions_to_Analytical_Bottlenecks/32627343"],"dc:rights":["All rights reserved","Open Access after 2026-12-15"],"dc:subject":["YOLO","BRUV","Deep learning","Behavioural analysis","Zebrafish","Artificial Intelligence","Fish detection","Object detection","Computer Vision","Underwater Video","Animal Tracking"],"dc:title":["Automating Video Analysis for Marine and Behavioural Research: Deep Learning Solutions to Analytical Bottlenecks"],"dc:type":["Text","Thesis"]},"updated_at":"2026-07-27T19:32:41Z"}