Impact of deep learning and post-processing algorithms performances on biodiversity metrics assessed on videos

PLOS One | https://doi.org/10.1371/journal.pone.0327577 August 11, 2025 1 / 14 OPEN ACCESS Citation: Fleuré V, Planolles K, Claverie T, Mulot B, Villéger S (2025) Impact of deep learning and post-processing algorithms performances on biodiversity metrics assessed on videos. PLoS One 20(8): e0327577. https:// doi.org/10.1371/journal.pone.0327577 Editor: Tzen-Yuh Chiang, National Cheng Kung University, TAIWAN Received: March 17, 2025 Accepted: June 17, 2025 Published: August 11, 2025 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal. pone.0327577 Copyright: © 2025 Fleuré et al . This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, RESEARCH ARTICLE Impact of deep learning and post-processing algorithms performances on biodiversity metrics assessed on videos Valentine Fleuré 1,2*, Kévin Planolles1,3, Thomas Claverie4, Baptiste Mulot2, Sébastien Villéger1 1 MARBEC, University Montpellier, CNRS, Ifremer, IRD, Montpellier, France, 2 ZooParc de Beauval & Beauval Nature, Saint-Aignan, France, 3 Research-team ICAR, LIRMM, University Montpellier, CNRS, Montpellier, France, 4 UMR ENTROPIE, IRD, IFREMER, CNRS, Univ La Réunion, Saint Denis, Réunion, France * valentine.fleure@gmail.com Abstract

Assessing the escalating biodiversity crisis, driven by climate change, habitat destruction, and exploitation, necessitates efficient monitoring strategies to assess species presence and abundance across diverse habitats. Video-based surveys using remote cameras are a promising, non-invasive way to collect valuable data in various environments. Yet, the analysis of recorded videos remains challenging due to time and expertise constraints. Recent advances in deep learning models have enhanced image processing capabilities in both object detection and classifi- cation. However, the impacts on models’ performances and usage on assessment of biodiversity metrics on videos is yet to be assessed. This study evaluates the impacts of video processing rates, detection and identification model performance, and post-processing algorithms on the accuracy of biodiversity metrics, using simu- lated remote videos of fish communities and 14,406 simulated automated processing pipelines. We found that a processing rate of one image per second minimizes errors while ensuring detection of all species. However, even near-perfect detection (both recall and precision of 0.99) and identification (accuracy of 0.99) models resulted in overestimation of total abundance, species richness and species diversity due to false positives. We reveal that post-processing model outputs using a confidence threshold approach (i.e., to discard most erroneous predictions while also discarding a smaller proportion of correct predictions) is the most efficient method to accurately estimate biodiversity from videos.

Introduction The escalating biodiversity crisis, exacerbated by climate change, habitat destruction, and exploitation, underscores the urgent need for efficient monitoring strategies of