@article{fdi:010093568,
  title = {{D}eep learning for automated coral reef monitoring a novel system based on {YOLO}v8 detection and {D}eep{SORT} tracking},
  author = {{O}uassine, {Y}. and {C}onruyt, {N}. and {K}ayal, {M}ohsen and {M}artin, {P}. {A}. and {B}igot, {L}. and {R}egine, {V}. {L}. and {M}oussanif, {H}. and {Z}ahir, {J}.},
  editor = {},
  language = {{ENG}},
  abstract = {{C}oral reefs are vital for biodiversity, coastal protection, food security, and tourism, yet they face severe threats from anthropogenic activities and climate change, which are leading to their decline. {E}ffective coral reef monitoring is essential for ecological understanding and conservation, but traditional methods are resourceintensive and rely on experts. {T}o address these challenges, we present an automated, deep learning-based monitoring system that integrates {YOLO}v8, a state-of-the-art object detection algorithm, with {D}eep{SORT}, a robust multi-object tracking method, to identify and track coral formations in underwater video footage. {O}ur system was fine-tuned using two curated and annotated datasets: {AIMECORAL}1 (580 images from the {S}outhwest {I}ndian {O}cean) and {AIMECORAL}2 (282 images from {N}ew {C}aledonia, {P}acific {O}cean), encompassing diverse coral species and environmental conditions. {T}he system's performance was evaluated using established metrics: object detection precision, {M}ultiple {O}bject {T}racking {A}ccuracy ({MOTA}), {M}ultiple {O}bject {T}racking {P}recision ({MOTP}), and {I}dentity {F}1 {S}core ({IDF}1). {P}recision improved from 59.9 % (after fine-tuning on {AIMECORAL}1) to 84.7 % on the combined datasets. {T}he tracking system achieved a {MOTA} of 82.63 %, {MOTP} of 83.28 %, and {IDF}1 of 70.76 %, demonstrating reliable multi-object tracking in complex underwater environments. {W}e applied our framework to a case study involving video transects from an outer reef site in {N}ew {C}aledonia, comparing data from 2021 and 2022. {T}his automated solution offers a scalable, cost-effective alternative to traditional monitoring methods, supporting seamless, large-scale reef assessment. {B}y leveraging deep learning, our approach enables more efficient data collection, contributing to the protection of these vulnerable ecosystems in the face of increasing environmental pressures.},
  keywords = {{C}oral reefs ecosystems ; {D}eep learning ; {M}ethodology ; {YOLO}v8 ; {D}eep{SORT}},
  booktitle = {},
  journal = {{E}cological {I}nformatics},
  volume = {89},
  numero = {},
  pages = {103170 [13 p.]},
  ISSN = {1574-9541},
  year = {2025},
  DOI = {10.1016/j.ecoinf.2025.103170},
  URL = {https://www.documentation.ird.fr/hor/fdi:010093568},
}