@article{fdi:010087462,
  title = {{D}etection of {S}argassum from {S}entinel {S}atellite {S}ensors {U}sing {D}eep {L}earning {A}pproach},
  author = {{L}aval, {M}. and {B}elmouhcine, {A}. and {C}ourtrai, {L}. and {D}escloitres, {J}. and {S}alazar-{G}aribay, {A}. and {S}chamberger, {L}. and {M}inghelli, {A}. and {T}hibaut, {T}. and {D}orville, {R}. and {M}azoyer, {C}. and {Z}ongo, {P}. and {C}hevalier, {C}rist{\`e}le},
  editor = {},
  language = {{ENG}},
  abstract = {{S}ince 2011, the proliferation of brown macro-algae of the genus {S}argassum has considerably increased in the {N}orth {T}ropical {A}tlantic {S}ea, all the way from the {G}ulf of {G}uinea to the {C}aribbean {S}ea and the {G}ulf of {M}exico. {T}he large amount of {S}argassum aggregations in that area cause major beaching events, which have a significant impact on the local economy and the environment and are starting to present a real threat to public health. {I}n such a context, it is crucial to collect spatial and temporal data of {S}argassum aggregations to understand their dynamics and predict stranding. {L}ately, indexes based on satellite imagery such as the {M}aximum {C}hlorophyll {I}ndex ({MCI}) or the {A}lternative {F}loating {A}lgae {I}ndex ({AFAI}), have been developed and used to detect these {S}argassum aggregations. {H}owever, their accuracy is questionable as they tend to detect various non-{S}argassum features. {T}o overcome false positive detection biases encountered by the index-thresholding methods, we developed two new deep learning models specific for {S}argassum detection based on an encoder-decoder convolutional neural network ({CNN}). {O}ne was tuned to spectral bands from the multispectral instrument ({MSI}) onboard {S}entinel-2 satellites and the other to the {O}cean and {L}and {C}olour {I}nstrument ({OLCI}) onboard {S}entinel-3 satellites. {T}his specific new approach outperformed previous generalist deep learning models, such as {E}ris{N}et, {UN}et, and {S}eg{N}et, in the detection of {S}argassum from satellite images with the same training, with an {F}1-score of 0.88 using {MSI} images, and 0.76 using {OLCI} images. {I}ndeed, the proposed {CNN} considered neighbor pixels, unlike {E}ris{N}et, and had fewer reduction levels than {UN}et and {S}eg{N}et, allowing filiform objects such as {S}argassum aggregations to be detected. {U}sing both spectral and spatial features, it also yielded a better detection performance compared to algal index-based techniques. {T}he {CNN} method proposed here recognizes new small aggregations that were previously undetected, provides more complete structures, and has a lower false-positive detection rate.},
  keywords = {ocean color ; {S}argassum ; {MODIS} ; {MSI} ; {OLCI} ; {S}entinel-2 ; {S}entinel-3 ; convolutional neural network ; deep learning ; {ATLANTIQUE} {NORD}},
  booktitle = {},
  journal = {{R}emote {S}ensing},
  volume = {15},
  numero = {4},
  pages = {1104 [21 ]},
  year = {2023},
  DOI = {10.3390/rs15041104},
  URL = {https://www.documentation.ird.fr/hor/fdi:010087462},
}