摘要
鼠害是我国草原的主要生物灾害之一,严重威胁草原畜牧业的可持续发展,利用神经网络模型能够实现复杂环境下草原地物类型的识别与分割,该方法在草原鼠害监测与管理方面具有重要意义。本研究以若尔盖草原为例,将Mask R-CNN卷积神经网络模型、 Res2Net网络与无人机影像数据相结合,提出了一种新型草原地物类型的识别与分割方法,对草原不同地物类型进行检测和分割,进而统计相应地物类型的面积与数量,建立了若尔盖草原的鼠害监测模型。得出如下结论:无人机低空遥感结合卷积神经网络模型的监测方法可以为若尔盖草原鼠害调查提供准确度甚高的解译结果;其中,土丘的检测精度最高,恢复斑块检测精度最低,但其面积占比最高;样地中土丘分布均匀,鼠洞呈集聚分布,两类害鼠分布空间、分布差异较大。研究结果反映的信息与鼠害发生区域基本一致,该方法为有效监测鼠害提供了决策支持,对有效保护草原生产力和实现草原的可持续发展具有重要的研究价值。
Grassland rodent not only seriously threatens the sustainable development of animal husbandry,but also exacerbates environmental degradation in China.In the past decades,unmanned aerial vehicle(UAV)has received much attention due to the advantages in efficiency,objectivity and timeliness.Besides,as a popular method convolutional neural network(CNN)plays an important role in monitoring and management of grassland,since CNN could detect and segment the different objects even in complicate circumstance.The objective of this study was to investigate the potential of CNN in monitoring grassland rodent and to examine the feasibility of developing universal models for the different experiment sites.The experiments were conducted in Ruoergai grassland with extensive image data collected at multiple phenological stages of grass growth across the entire season with UAV.The rodent monitoring model established by coupling Mask R-CNN with UAV images was used to detect four classification in grassland and calculate these area and quantity.The results suggest coupling UAV with CNN is a promising approach to monitoring rodent condition.Mound was exhibited higher detection accuracies among the four classification,which were mound,bare land,recuperative patch and mousehole.Recuperative plaque with the lowest detection accuracies takes the most proportion.Further analysis revealed that the mounds are evenly distributed and the mouse holes are clustered.The spatial distribution of the two types of pests is quite different.Among the four types of land,the area of ecological restoration patches accounted for the highest proportion,indicating that the destroyed grassland vegetation is currently in the early stage of ecological restoration and grassland protection has achieved certain results.The account of mound in different experiment sites were similar.Compared to mound,the account of mousehole were exhibited remarkably difference in various experiment sites.The results derived from research are consistent with the real condition in
作者
周俗
韩立亮
杨思维
王钰
根呷羊批
牛培莉
王泽光
Zhou Su;Han Liliang;Yang Siwei;Wang Yu;GenXia Yangpi;Niu Peili;Wang Zeguang(Sichuan Academy of Grassland Sciences/Institute of Alpine Grassland Ecological Restoration Engineering technology on Qinghai-Tibet Plateau,Chengdu 611743,China;Tsinghua-QingDao Big Data Engineering Research Center,Qingdao 266000,China;Southwest university for nationalities institute of Qinghai-Tibet plateau,Chengdu 610041,China;Aba Prefecture Forestry and Grassland Administration,Maerkang 624000,China;Ruoergai County Forestry and Grassland Bureau Grassland Supervision Station,Ruoergai 624500,China)
出处
《草学》
2021年第2期15-25,共11页
Journal of Grassland and Forage Science
基金
国家重点研发计划项目(2017YFC0504803)
四川省科技厅项目“高寒草地高原鼢鼠栖息地选择机制研究”(2019YJ0569)
西南民族大学研究生创新型科研项目(CX2019SZ85)
青藏高原草地鼠害控制新技术效果评价(2018CC0137-2)
国家现代农业产业技术体系四川饲草创新团队专项建设资金(sccxtd-2020-16)共同资助。
