@article{fdi:010087745,
  title = {{A}ccurate delineation of individual tree crowns in tropical forests from aerial {RGB} imagery using {M}ask {R}-{CNN}},
  author = {{B}all, {J}. {G}. {C}. and {H}ickman, {S}. {H}. {M}. and {J}ackson, {T}. {D}. and {K}oay, {X}. {J}. and {H}irst, {J}. and {J}ay, {W}. and {A}rcher, {M}. and {A}ubry-{K}ientz, {M}. and {V}incent, {G}r{\'e}goire and {C}oomes, {D}. {A}.},
  editor = {},
  language = {{ENG}},
  abstract = {{T}ropical forests are a major component of the global carbon cycle and home to two-thirds of terrestrial species. {U}pper-canopy trees store the majority of forest carbon and can be vulnerable to drought events and storms. {M}onitoring their growth and mortality is essential to understanding forest resilience to climate change, but in the context of forest carbon storage, large trees are underrepresented in traditional field surveys, so estimates are poorly constrained. {A}erial photographs provide spectral and textural information to discriminate between tree crowns in diverse, complex tropical canopies, potentially opening the door to landscape monitoring of large trees. {H}ere we describe a new deep convolutional neural network method, {D}etectree2, which builds on the {M}ask {R}-{CNN} computer vision framework to recognize the irregular edges of individual tree crowns from airborne {RGB} imagery. {W}e trained and evaluated this model with 3797 manually delineated tree crowns at three sites in {M}alaysian {B}orneo and one site in {F}rench {G}uiana. {A}s an example application, we combined the delineations with repeat lidar surveys (taken between 3 and 6 years apart) of the four sites to estimate the growth and mortality of upper-canopy trees. {D}etectree2 delineated 65 000 upper-canopy trees across 14 km(2) of aerial images. {T}he skill of the automatic method in delineating unseen test trees was good ({F}-1 score = 0.64) and for the tallest category of trees was excellent ({F}-1 score = 0.74). {A}s predicted from previous field studies, we found that growth rate declined with tree height and tall trees had higher mortality rates than intermediate-size trees. {O}ur approach demonstrates that deep learning methods can automatically segment trees in widely accessible {RGB} imagery. {T}his tool (provided as an open-source {P}ython package) has many potential applications in forest ecology and conservation, from estimating carbon stocks to monitoring forest phenology and restoration.{P}ython package available to install at https://github.com/{P}at{B}all1/{D}etectree2 .},
  keywords = {{C}onvolutional neural networks ; deep learning ; {D}etectron2 ; forest monitoring ; {M}ask {R}-{CNN} ; tree crown delineation ; tree crown segmentation ; tree growth ; tree mortality ; tropical forests ; {MALAISIE} ; {GUYANE} {FRANCAISE} ; {ZONE} {TROPICALE}},
  booktitle = {},
  journal = {{R}emote {S}ensing in {E}cology and {C}onservation},
  volume = {[{E}arly access]},
  numero = {},
  pages = {[14 ]},
  year = {2023},
  DOI = {10.1002/rse2.332},
  URL = {https://www.documentation.ird.fr/hor/fdi:010087745},
}