Deep Learning for Dynamic Wildlife Monitoring: A Real-Time Approach

<p dir="ltr">With the rapid amount of deforestation around the world, wild animals lose their habitat and move closer to human settlements, leading to increased human-wildlife conflict and the loss of important human lives, livestock, and rare wild animal species. Wildlife monitoring...

Full description

Saved in:
Bibliographic Details
Main Author: Abdul Basit Mughal (22929001) (author)
Other Authors: Rafi Ullah Khan (22929004) (author), Atiq Ur Rehman (8843024) (author), and Amine Bermak (21400739) (author)
Published: 2025
Subjects:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1864513525463908352
author Abdul Basit Mughal (22929001)
author2 Rafi Ullah Khan (22929004)
Atiq Ur Rehman (8843024)
and Amine Bermak (21400739)
author2_role author
author
author
author_facet Abdul Basit Mughal (22929001)
Rafi Ullah Khan (22929004)
Atiq Ur Rehman (8843024)
and Amine Bermak (21400739)
author_role author
dc.creator.none.fl_str_mv Abdul Basit Mughal (22929001)
Rafi Ullah Khan (22929004)
Atiq Ur Rehman (8843024)
and Amine Bermak (21400739)
dc.date.none.fl_str_mv 2025-08-26T12:00:00Z
dc.identifier.none.fl_str_mv 10.1109/access.2025.3600625
dc.relation.none.fl_str_mv https://figshare.com/articles/journal_contribution/Deep_Learning_for_Dynamic_Wildlife_Monitoring_A_Real-Time_Approach/30971755
dc.rights.none.fl_str_mv CC BY 4.0
info:eu-repo/semantics/openAccess
dc.subject.none.fl_str_mv Environmental sciences
Environmental management
Information and computing sciences
Computer vision and multimedia computation
Machine learning
Wildlife
object detection
wild-life monitoring
computer vision
deep learning
Animals
Monitoring
Biological system modeling
YOLO
Computational modeling
Accuracy
Feature extraction
Detectors
Cameras
dc.title.none.fl_str_mv Deep Learning for Dynamic Wildlife Monitoring: A Real-Time Approach
dc.type.none.fl_str_mv Text
Journal contribution
info:eu-repo/semantics/publishedVersion
text
contribution to journal
description <p dir="ltr">With the rapid amount of deforestation around the world, wild animals lose their habitat and move closer to human settlements, leading to increased human-wildlife conflict and the loss of important human lives, livestock, and rare wild animal species. Wildlife monitoring is an essential task in computer vision for maintaining biodiversity, ecosystem balance and stability. It plays a vital role where traditional systems, such as camera traps and manual methods, fall short, and helps to avoid human-wildlife conflict. However, due to the challenging forest terrains, dynamic backgrounds, diverse animal poses, and low lighting conditions, existing methods struggle to efficiently extract features in real-time. These models suffer from limitations, including slow speed and low accuracy. To address these challenges, a novel real-time wildlife monitoring model has been introduced that leverages YOLOv11, designed to perform effectively in complex backgrounds, diverse animal poses and low-light conditions. A new wild animal dataset is developed, comprising 17 different wild animal species, captured under low-light conditions and in challenging backgrounds. The Grad-CAM technique is integrated with YOLOv11 to enhance the model’s interpretability to generate heatmaps and attention maps. These maps help identify the important regions of the animal’s body on which the model focuses more in real-time monitoring applications. The YOLOv11n model achieves an mAP of 97.4%, precision of 96%, and an F1-score of 95% on self-created wild animal dataset. The parameters of the fine-tuned YOLOv11 decreased by 30% relative to the previous version, resulting in a very small model size of 5.5 MB and reduced processing time. The model was evaluated on two other datasets: Animal2 and Camera Trap, with an mAP of 92% on the Animal2 dataset and 90% on the Camera Trap dataset.</p><h2 dir="ltr">Other Information</h2><p dir="ltr">Published in: IEEE Access<br>License: <a href="https://creativecommons.org/licenses/by/4.0/deed.en" target="_blank">https://creativecommons.org/licenses/by/4.0/</a><br>See article on publisher's website: <a href="https://dx.doi.org/10.1109/access.2025.3600625" target="_blank">https://dx.doi.org/10.1109/access.2025.3600625</a></p>
eu_rights_str_mv openAccess
id Manara2_447204f52bbe7737c0c55e02efe86ca4
identifier_str_mv 10.1109/access.2025.3600625
network_acronym_str Manara2
network_name_str Manara2
oai_identifier_str oai:figshare.com:article/30971755
publishDate 2025
repository.mail.fl_str_mv
repository.name.fl_str_mv
repository_id_str
rights_invalid_str_mv CC BY 4.0
spelling Deep Learning for Dynamic Wildlife Monitoring: A Real-Time ApproachAbdul Basit Mughal (22929001)Rafi Ullah Khan (22929004)Atiq Ur Rehman (8843024)and Amine Bermak (21400739)Environmental sciencesEnvironmental managementInformation and computing sciencesComputer vision and multimedia computationMachine learningWildlifeobject detectionwild-life monitoringcomputer visiondeep learningAnimalsMonitoringBiological system modelingYOLOComputational modelingAccuracyFeature extractionDetectorsCameras<p dir="ltr">With the rapid amount of deforestation around the world, wild animals lose their habitat and move closer to human settlements, leading to increased human-wildlife conflict and the loss of important human lives, livestock, and rare wild animal species. Wildlife monitoring is an essential task in computer vision for maintaining biodiversity, ecosystem balance and stability. It plays a vital role where traditional systems, such as camera traps and manual methods, fall short, and helps to avoid human-wildlife conflict. However, due to the challenging forest terrains, dynamic backgrounds, diverse animal poses, and low lighting conditions, existing methods struggle to efficiently extract features in real-time. These models suffer from limitations, including slow speed and low accuracy. To address these challenges, a novel real-time wildlife monitoring model has been introduced that leverages YOLOv11, designed to perform effectively in complex backgrounds, diverse animal poses and low-light conditions. A new wild animal dataset is developed, comprising 17 different wild animal species, captured under low-light conditions and in challenging backgrounds. The Grad-CAM technique is integrated with YOLOv11 to enhance the model’s interpretability to generate heatmaps and attention maps. These maps help identify the important regions of the animal’s body on which the model focuses more in real-time monitoring applications. The YOLOv11n model achieves an mAP of 97.4%, precision of 96%, and an F1-score of 95% on self-created wild animal dataset. The parameters of the fine-tuned YOLOv11 decreased by 30% relative to the previous version, resulting in a very small model size of 5.5 MB and reduced processing time. The model was evaluated on two other datasets: Animal2 and Camera Trap, with an mAP of 92% on the Animal2 dataset and 90% on the Camera Trap dataset.</p><h2 dir="ltr">Other Information</h2><p dir="ltr">Published in: IEEE Access<br>License: <a href="https://creativecommons.org/licenses/by/4.0/deed.en" target="_blank">https://creativecommons.org/licenses/by/4.0/</a><br>See article on publisher's website: <a href="https://dx.doi.org/10.1109/access.2025.3600625" target="_blank">https://dx.doi.org/10.1109/access.2025.3600625</a></p>2025-08-26T12:00:00ZTextJournal contributioninfo:eu-repo/semantics/publishedVersiontextcontribution to journal10.1109/access.2025.3600625https://figshare.com/articles/journal_contribution/Deep_Learning_for_Dynamic_Wildlife_Monitoring_A_Real-Time_Approach/30971755CC BY 4.0info:eu-repo/semantics/openAccessoai:figshare.com:article/309717552025-08-26T12:00:00Z
spellingShingle Deep Learning for Dynamic Wildlife Monitoring: A Real-Time Approach
Abdul Basit Mughal (22929001)
Environmental sciences
Environmental management
Information and computing sciences
Computer vision and multimedia computation
Machine learning
Wildlife
object detection
wild-life monitoring
computer vision
deep learning
Animals
Monitoring
Biological system modeling
YOLO
Computational modeling
Accuracy
Feature extraction
Detectors
Cameras
status_str publishedVersion
title Deep Learning for Dynamic Wildlife Monitoring: A Real-Time Approach
title_full Deep Learning for Dynamic Wildlife Monitoring: A Real-Time Approach
title_fullStr Deep Learning for Dynamic Wildlife Monitoring: A Real-Time Approach
title_full_unstemmed Deep Learning for Dynamic Wildlife Monitoring: A Real-Time Approach
title_short Deep Learning for Dynamic Wildlife Monitoring: A Real-Time Approach
title_sort Deep Learning for Dynamic Wildlife Monitoring: A Real-Time Approach
topic Environmental sciences
Environmental management
Information and computing sciences
Computer vision and multimedia computation
Machine learning
Wildlife
object detection
wild-life monitoring
computer vision
deep learning
Animals
Monitoring
Biological system modeling
YOLO
Computational modeling
Accuracy
Feature extraction
Detectors
Cameras