@article{fdi:010092807,
  title = {{A}utomatic detection, identification and counting of deep-water snappers on underwater baited video using deep learning},
  author = {{B}aletaud, {F}. and {V}illon, {S}{\'e}bastien and {G}ilbert, {A}. and {C}{\^o}me, {J}. {M}. and {F}iat, {S}ylvie and {I}ovan, {C}orina and {V}igliola, {L}aurent},
  editor = {},
  language = {{ENG}},
  abstract = {{D}eep-sea demersal fisheries in the {P}acific have strong commercial, cultural, and recreational value, especially snappers ({L}utjanidae) which make the bulk of catches. {Y}et, managing these fisheries is challenging due to the scarcity of data. {S}tereo-{B}aited {R}emote {U}nderwater {V}ideo {S}tations ({BRUVS}) can provide valuable quantitative information on fish stocks, but manually processing large amounts of videos is time-consuming and sometimes unrealistic. {T}o address this issue, we used a {R}egion-based {C}onvolutional {N}eural {N}etwork ({F}aster {R}-{CNN}), a deep learning architecture to automatically detect, identify and count deep-water snappers in {BRUVS}. {V}ideos were collected in {N}ew {C}aledonia ({S}outh {P}acific) at depths ranging from 47 to 552 m. {U}sing a dataset of 12,100 annotations from 11 deep-water snapper species observed in 6,364 images, we obtained good model performance for the 6 species with sufficient annotations ({F}-measures >0.7, up to 0.87). {T}he correlation between automatic and manual estimates of fish {M}ax{N} abundance in videos was high (0.72 - 0.9), but the {F}aster {R}-{CNN} showed an underestimation bias at higher abundances. {A} semi-automatic protocol where our model supported manual observers in processing {BRUVS} footage improved performance with a correlation of 0.96 with manual counts and a perfect match ({R}=1) for some key species. {T}his model can already assist manual observers to semi-automatically process {BRUVS} footage and will certainly improve when more training data will be available to decrease the rate of false negatives. {T}his study further shows that the use of artificial intelligence in marine science is progressive but warranted for the future.},
  keywords = {deep-water snapper fisheries ; artificial intelligence ; semi-automatic ; {BRUVS} ; faster {R}-{CNN} ; {NOUVELLE} {CALEDONIE} ; {PACIFIQUE} {SUD}},
  booktitle = {},
  journal = {{F}rontiers in {M}arine {S}cience},
  volume = {12},
  numero = {},
  pages = {1476616 [ p.]},
  year = {2025},
  DOI = {10.3389/fmars.2025.1476616},
  URL = {https://www.documentation.ird.fr/hor/fdi:010092807},
}