@article{fdi:010082188,
  title = {{A}ssessment of deep learning techniques for land use land cover classification in {S}outhern {N}ew {C}aledonia},
  author = {{R}ousset, {G}. and {D}espinoy, {M}arc and {S}chindler, {K}. and {M}angeas, {M}organ},
  editor = {},
  language = {{ENG}},
  abstract = {{L}and use ({LU}) and land cover ({LC}) are two complementary pieces of cartographic information used for urban planning and environmental monitoring. {I}n the context of {N}ew {C}aledonia, a biodiversity hotspot, the availability of up-to-date {LULC} maps is essential to monitor the impact of extreme events such as cyclones and human activities on the environment. {W}ith the democratization of satellite data and the development of high-performance deep learning techniques, it is possible to create these data automatically. {T}his work aims at determining the best current deep learning configuration (pixel-wise vs. semantic labelling architectures, data augmentation, image prepossessing, horizontal ellipsis ), to perform {LULC} mapping in a complex, subtropical environment. {F}or this purpose, a specific data set based on {SPOT}6 satellite data was created and made available for the scientific community as an {LULC} benchmark in a tropical, complex environment using five representative areas of {N}ew {C}aledonia labelled by a human operator: four used as training sets, and the fifth as a test set. {S}everal architectures were trained and the resulting classification was compared with a state-of-the-art machine learning technique: {XG}boost. {W}e also assessed the relevance of popular neo-channels derived from the raw observations in the context of deep learning. {T}he deep learning approach showed comparable results to {XG}boost for {LC} detection and over-performed it on the {LU} detection task (61.45% vs. 51.56% of overall accuracy). {F}inally, adding {LC} classification output of the dedicated deep learning architecture to the raw channels input significantly improved the overall accuracy of the deep learning {LU} classification task (63.61% of overall accuracy). {A}ll the data used in this study are available on line for the remote sensing community and for assessing other {LULC} detection techniques.},
  keywords = {{N}ew {C}aledonia ; remote sensing ; land use ; land cover ; deep learning ; {XGB}oost ; neural network ; neo-channels ; {NOUVELLE} {CALEDONIE}},
  booktitle = {},
  journal = {{R}emote {S}ensing},
  volume = {13},
  numero = {12},
  pages = {2257 [22 ]},
  year = {2021},
  DOI = {10.3390/rs13122257},
  URL = {https://www.documentation.ird.fr/hor/fdi:010082188},
}