Deep learning‐based ecological analysis of camera trap images is impacted by training data quality and quantity
Remote Sensing in Ecology and Conservation · 2026-06 · 12 (3) · 494-508
DOI: 10.1002/rse2.70052 ↗Kurz gefasst
Fachartikel in der Zeitschrift Remote Sensing in Ecology and Conservation, online erschienen am 11.01.2026. Verfasst von Peggy A. Bevan und 12 weiteren Autor:innen. Die Arbeit wurde bisher 6-mal zitiert (Crossref) und nennt 69 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Natural Environment Research Council, UK Research and Innovation.
Abstract (Originalsprache)
Large image collections generated from camera traps offer valuable insights into species richness, occupancy, and activity patterns, significantly aiding biodiversity monitoring. However, the manual processing of these data sets is time‐consuming, hindering analytical processes. To address this, deep neural networks have been widely adopted to automate image labelling, but the impact of classification error on key ecological metrics remains unclear. Here, we analyze data from camera trap collections in an African savannah (82,300 labelled images, 47 species) and an Asian sub‐tropical dry forest (40,308 labelled images, 29 species) to compare ecological metrics derived from expert‐generated species identifications with those generated by deep‐learning classification models. We specifically assess the impact of deep‐ learning model architecture, the proportion of label noise in the training data, and the size of the training data set on three key ecological metrics: species richness, occupancy, and activity patterns. We found that predictions of species richness derived from deep neural networks closely match those calculated from expert labels and remained resilient to up to 10% noise in the training data set (mis‐labelled images) and a 50% reduction in the training data set size. We found that our choice of deep‐learning model architecture (ResNet vs. ConvNext‐T) or depth (ResNet18, 50, 101) did not impact predicted ecological metrics. In contrast, species‐specific metrics were more sensitive; less common and visually similar species were disproportionately affected by a reduction in deep neural network accuracy, with consequences for occupancy and diel activity pattern estimates. To ensure the reliability of their findings, practitioners should prioritize creating large, clean training sets and account for class imbalance across species over exploring numerous deep‐learning model architectures.
Förderung
- Natural Environment Research Council
- UK Research and Innovation
Zitieren
APA
Bevan, P. A., Pantazis, O., Pringle, H. A., Ferreira, G. B., Ingram, D. J., Madsen, E. K., Thomas, L., Thanet, D. R., Silwal, T., Rayamajhi, S., Brostow, G. J., Aodha, O. M., & Jones, K. E. (2026). Deep learning‐based ecological analysis of camera trap images is impacted by training data quality and quantity. Remote Sensing in Ecology and Conservation, 12(3), 494-508. https://doi.org/10.1002/rse2.70052
Vancouver
Bevan PA, Pantazis O, Pringle HA, Ferreira GB, Ingram DJ, Madsen EK, et al. Deep learning‐based ecological analysis of camera trap images is impacted by training data quality and quantity. Remote Sensing in Ecology and Conservation. 2026;12(3):494-508. doi:10.1002/rse2.70052
BibTeX
@article{Bevan2026,
author = {Bevan, Peggy A. and Pantazis, Omiros and Pringle, Holly A.I. and Ferreira, Guilherme Braga and Ingram, Daniel J. and Madsen, Emily K. and Thomas, Liam and Thanet, Dol Raj and Silwal, Thakur and Rayamajhi, Santosh and Brostow, Gabriel J. and Aodha, Oisin Mac and Jones, Kate E.},
title = {Deep learning‐based ecological analysis of camera trap images is impacted by training data quality and quantity},
journal = {Remote Sensing in Ecology and Conservation},
year = {2026},
volume = {12},
number = {3},
pages = {494--508},
doi = {10.1002/rse2.70052}
}
EUVIMED