Solution for diagnostics of biological invasion in terrestrial ecosystems: how can deep learning help biodiversity conservation?

IF 2.5 3区环境科学与生态学 Q2 BIODIVERSITY CONSERVATION

Journal for Nature Conservation Pub Date : 2025-08-26 DOI:10.1016/j.jnc.2025.127066

Giovanna de Andrade Ferreira , José Matheus Segre Moneva Viveiros , Giulio Brossi Santoro , Vinicius Cunha Amaral , Matheus Pinheiro Ferreira , Pedro Henrique Santin Brancalion , Paulo Guilherme Molin

{"title":"Solution for diagnostics of biological invasion in terrestrial ecosystems: how can deep learning help biodiversity conservation?","authors":"Giovanna de Andrade Ferreira , José Matheus Segre Moneva Viveiros , Giulio Brossi Santoro , Vinicius Cunha Amaral , Matheus Pinheiro Ferreira , Pedro Henrique Santin Brancalion , Paulo Guilherme Molin","doi":"10.1016/j.jnc.2025.127066","DOIUrl":null,"url":null,"abstract":"<div><div>Tree invasions are a serious threat to grassland ecosystems, but control measures often rely on diagnostic approaches that are not yet effective or scalable. This study applied the deep learning algorithm Mask R-CNN to detect an invasive exotic species (<em>Pinus elliottii</em>) in a native wetland area, aiming to create a biological invasion diagnostic tool to support the management of native areas and assist in biological invasion control. The model was developed using high spatial resolution images (1.5 cm/pixel) and achieved a mean Average Precision (mAP) of 78 % and an Intersection over Union (IoU) score of 81 %. The segmentations generated by the model provides an assessment of the biological invasion process caused by <em>Pinus</em> spp. through the detection of individual trees, quantify the canopy cover area affected and evaluate the effectiveness of the method as a supporting tool for biodiversity conservation in protected areas. Recognizing the management priority of invasive exotic species and the limitations of available tools for public managers, the deep learning approach presented here may contribute to the development of diagnostics that inform more targeted and effective management actions, reducing financial costs, environmental impacts, and time spent on field activities.</div></div>","PeriodicalId":54898,"journal":{"name":"Journal for Nature Conservation","volume":"89 ","pages":"Article 127066"},"PeriodicalIF":2.5000,"publicationDate":"2025-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal for Nature Conservation","FirstCategoryId":"93","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S1617138125002432","RegionNum":3,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"BIODIVERSITY CONSERVATION","Score":null,"Total":0}

引用次数: 0

Abstract

Tree invasions are a serious threat to grassland ecosystems, but control measures often rely on diagnostic approaches that are not yet effective or scalable. This study applied the deep learning algorithm Mask R-CNN to detect an invasive exotic species (Pinus elliottii) in a native wetland area, aiming to create a biological invasion diagnostic tool to support the management of native areas and assist in biological invasion control. The model was developed using high spatial resolution images (1.5 cm/pixel) and achieved a mean Average Precision (mAP) of 78 % and an Intersection over Union (IoU) score of 81 %. The segmentations generated by the model provides an assessment of the biological invasion process caused by Pinus spp. through the detection of individual trees, quantify the canopy cover area affected and evaluate the effectiveness of the method as a supporting tool for biodiversity conservation in protected areas. Recognizing the management priority of invasive exotic species and the limitations of available tools for public managers, the deep learning approach presented here may contribute to the development of diagnostics that inform more targeted and effective management actions, reducing financial costs, environmental impacts, and time spent on field activities.

查看原文本刊更多论文

陆地生态系统生物入侵诊断的解决方案：深度学习如何帮助生物多样性保护？

树木入侵是对草地生态系统的严重威胁，但控制措施往往依赖于尚未有效或可扩展的诊断方法。本研究应用深度学习算法Mask R-CNN对原生湿地外来入侵种（Pinus elliottii）进行检测，旨在创建一种生物入侵诊断工具，为原生湿地管理提供支持，协助生物入侵控制。该模型使用高空间分辨率图像（1.5 cm/像素）开发，平均平均精度（mAP）为78%，交汇（IoU）评分为81%。该模型生成的分割结果可通过单株检测来评估松木的生物入侵过程，量化受影响的冠层覆盖面积，并评价该方法作为保护区生物多样性保护的辅助工具的有效性。认识到外来入侵物种管理的优先性和公共管理者可用工具的局限性，本文提出的深度学习方法可能有助于诊断的发展，为更有针对性和更有效的管理行动提供信息，减少财务成本、环境影响和现场活动花费的时间。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Journal for Nature Conservation 环境科学-生态学

CiteScore

3.70

自引率

5.00%

发文量

151

审稿时长

7.9 weeks

期刊介绍： The Journal for Nature Conservation addresses concepts, methods and techniques for nature conservation. This international and interdisciplinary journal encourages collaboration between scientists and practitioners, including the integration of biodiversity issues with social and economic concepts. Therefore, conceptual, technical and methodological papers, as well as reviews, research papers, and short communications are welcomed from a wide range of disciplines, including theoretical ecology, landscape ecology, restoration ecology, ecological modelling, and others, provided that there is a clear connection and immediate relevance to nature conservation. Manuscripts without any immediate conservation context, such as inventories, distribution modelling, genetic studies, animal behaviour, plant physiology, will not be considered for this journal; though such data may be useful for conservationists and managers in the future, this is outside of the current scope of the journal.