为解决由于无人机视角下毛竹林的形状和纹理复杂,现有方法在分割精度和鲁棒性方面表现不佳的问题,提出了一种应用跨领域适应和偏移量引导的毛竹林分割网络——BFSNet。以百山祖国家公园为试验区,利用无人机拍摄周边毛竹林图像构建数据...为解决由于无人机视角下毛竹林的形状和纹理复杂,现有方法在分割精度和鲁棒性方面表现不佳的问题,提出了一种应用跨领域适应和偏移量引导的毛竹林分割网络——BFSNet。以百山祖国家公园为试验区,利用无人机拍摄周边毛竹林图像构建数据集。为增强模型的特征提取能力,提出跨领域适应模块以有效利用源模型的强特征提取能力,并结合自主学习提取适用于毛竹林分割任务的特征,利用两者的优势进行互补。为提高模型对于不同形状毛竹林的识别和定位能力,结合可变形卷积的偏移量引导模块,引入可学习的偏移量参数,以适应不同形状的毛竹林目标。将BFSNet在DeepGlobe Land Cover Classification Challenge和自制数据集上进行模型训练和测试,并与多种主流图像分割方法进行对比。结果表明:BFSNet在交并比、Dice系数、精确率和召回率4项指标上均取得了最优的性能表现,分别获得了76.04%和71.93%的交并比。与多种主流的图像分割模型相比,BFSNet在毛竹林的分割效果方面表现最为出色,对毛竹林形状的精确建模能力能够有效地应对不同形态的毛竹林。展开更多
Remote sensing plays a pivotal role in forest inventory by enabling efficient large-scale monitoring while minimizing fieldwork costs.However,missing values pose a critical challenge in remote sensing applications,as ...Remote sensing plays a pivotal role in forest inventory by enabling efficient large-scale monitoring while minimizing fieldwork costs.However,missing values pose a critical challenge in remote sensing applications,as ignoring or mishandling such data gaps can introduce systematic bias into the estimation of target variables for natural resource monitoring.This can lead to cascading errors that propagate through forest and ecosystem management decisions,ultimately hindering progress toward sustainable forest management,biodiversity conservation,and climate change mitigation strategies.This study aims to propose and demonstrate a procedure that employs hybrid estimators to address the limitations of missing remotely sensed data in forest inventory,using Landsat 7 ETM+SLC-off data as an archived source for forest resource monitoring as a case in point.We compared forest inventory estimates from the hybrid estimator with those from a conventional model-based(CMB)estimator using Sentinel-2 data without missing values.Monte Carlo simulations revealed three key findings:(1)The hybrid estimator,leveraging missing-data remote sensing represented by Landsat 7 ETM+SLCoff data,achieved a sampling precision of over 90%,meeting China's national standard for the National Forest Inventory(NFI);(2)The hybrid estimator demonstrated comparable efficiency to the CMB estimator;(3)The uncertainty associated with hybrid estimators was primarily dominated by model parameter estimation,which could be effectively mitigated by slightly increasing the training sample size or refining model specification.Overall,in forest inventory,the hybrid estimator can surmount the limitations posed by missing values in remotely sensed auxiliary data,effectively balancing cost-effectiveness and flexibility.展开更多
文摘为解决由于无人机视角下毛竹林的形状和纹理复杂,现有方法在分割精度和鲁棒性方面表现不佳的问题,提出了一种应用跨领域适应和偏移量引导的毛竹林分割网络——BFSNet。以百山祖国家公园为试验区,利用无人机拍摄周边毛竹林图像构建数据集。为增强模型的特征提取能力,提出跨领域适应模块以有效利用源模型的强特征提取能力,并结合自主学习提取适用于毛竹林分割任务的特征,利用两者的优势进行互补。为提高模型对于不同形状毛竹林的识别和定位能力,结合可变形卷积的偏移量引导模块,引入可学习的偏移量参数,以适应不同形状的毛竹林目标。将BFSNet在DeepGlobe Land Cover Classification Challenge和自制数据集上进行模型训练和测试,并与多种主流图像分割方法进行对比。结果表明:BFSNet在交并比、Dice系数、精确率和召回率4项指标上均取得了最优的性能表现,分别获得了76.04%和71.93%的交并比。与多种主流的图像分割模型相比,BFSNet在毛竹林的分割效果方面表现最为出色,对毛竹林形状的精确建模能力能够有效地应对不同形态的毛竹林。
基金supported by the National Key R&D Program of China(No.2023YFF1304002-05)the National Social Science Fund of China(No.22BTJ005)the National Natural Science Foundation of China(No.32572049)。
文摘Remote sensing plays a pivotal role in forest inventory by enabling efficient large-scale monitoring while minimizing fieldwork costs.However,missing values pose a critical challenge in remote sensing applications,as ignoring or mishandling such data gaps can introduce systematic bias into the estimation of target variables for natural resource monitoring.This can lead to cascading errors that propagate through forest and ecosystem management decisions,ultimately hindering progress toward sustainable forest management,biodiversity conservation,and climate change mitigation strategies.This study aims to propose and demonstrate a procedure that employs hybrid estimators to address the limitations of missing remotely sensed data in forest inventory,using Landsat 7 ETM+SLC-off data as an archived source for forest resource monitoring as a case in point.We compared forest inventory estimates from the hybrid estimator with those from a conventional model-based(CMB)estimator using Sentinel-2 data without missing values.Monte Carlo simulations revealed three key findings:(1)The hybrid estimator,leveraging missing-data remote sensing represented by Landsat 7 ETM+SLCoff data,achieved a sampling precision of over 90%,meeting China's national standard for the National Forest Inventory(NFI);(2)The hybrid estimator demonstrated comparable efficiency to the CMB estimator;(3)The uncertainty associated with hybrid estimators was primarily dominated by model parameter estimation,which could be effectively mitigated by slightly increasing the training sample size or refining model specification.Overall,in forest inventory,the hybrid estimator can surmount the limitations posed by missing values in remotely sensed auxiliary data,effectively balancing cost-effectiveness and flexibility.
