Background： After their death, Scots pine trees can remain standing for decades and sometimes up to 200 years,forming long-lasting and ecologically important structures in boreal forest landscapes. Standing dead pine...Background： After their death, Scots pine trees can remain standing for decades and sometimes up to 200 years,forming long-lasting and ecologically important structures in boreal forest landscapes. Standing dead pines decay very slowly and with time develop into ‘kelo＇ trees, which are characterized by hard wood with silvery-colored appearance. These kelo trees represent an ecologically important, long lasting and visually striking element of the structure of natural pine-dominated forests in boreal Fennoscandia that is nowadays virtually absent from managed forest landscapes.Methods： We examined and mapped the amount, structural features, site characteristics and spatial distribution of dead standing pine trees over a ten hectare area in an unmanaged boreal forest landscape in the Kalevala National Park in Russian Viena Karelia.Results： The mean basal area of dead standing pine trees in the forested part of the landscape was 1.7 m^2·ha^-1 and the estimated volume 12.7 m^3·ha^-1. From the total number of standing dead pine trees 65% were kelo trees, with a basal area of 1.1 m^2·ha^-1 and volume of 8.0 m^3·ha^-1, the remainder consisting of standing dead pines along the continuum between a recently dead tree and a kelo tree. Overall, standing dead pines were distributed throughout the study area, but there was a tendency towards spatial clustering up to 〈100 m distances. Standing dead pines were most commonly situated on flat ground or in the mid slope in the local topography.In addition, standing dead pines contributed to substrate diversity also by commonly having charred wood and broken tops. Based on the presence of dead pine snags in different stage of transition from a recently dead pine to a kelo with silvery surface, it seems evident that the process of kelo recruitment was continuously in action in the studied landscape.Conclusions： Kelo trees are an omnipresent feature in natural pine-dominated forest landscapes with important contribution to forest structural and substrate diversity. Because of their longevity and extremely slow turnover dynamics and importance for biodiversity, protection of vulnerable kelo tree populations, and ensuring their continuous recruitment, should be of high priority in forest restoration and sustainable management.展开更多
为解决山地地形起伏大、无人机飞行高度高导致图像中尺度小且纹理模糊的松枯死木识别困难问题,该研究提出了一种在特征层级进行超分辨率重建的YOLOv5松枯死木识别算法。在YOLOv5网络中添加选择性核特征纹理迁移模块生成有细节纹理的高...为解决山地地形起伏大、无人机飞行高度高导致图像中尺度小且纹理模糊的松枯死木识别困难问题,该研究提出了一种在特征层级进行超分辨率重建的YOLOv5松枯死木识别算法。在YOLOv5网络中添加选择性核特征纹理迁移模块生成有细节纹理的高清检测特征图,自适应改变感受野的机制分配权重,将更多注意力集中在纹理细节,提升了小目标和模糊目标的识别精度。同时,使用前景背景平衡损失函数抑制背景噪声干扰,增加正样本的梯度贡献,改善正负样本分布不平衡问题。试验结果表明,改进后算法在交并比(intersection over union,IoU)阈值取0.5时的平均精度均值(mean average precision,mAP50)为92.7%,mAP50~95(以步长0.05从0.5到0.95间取IoU阈值下的平均mAP)为62.1%,APsmall(小目标平均精度值)为53.2%,相比于原算法mAP50提高了3.2个百分点,mAP50~95提升了8.3个百分点,APsmall提升了15.8个百分点。不同算法对比分析表明,该方法优于Faster R-CNN、YOLOv4、YOLOX、MT-YOLOv6,QueryDet、DDYOLOv5等深度学习算法,mAP50分别提高了16.7、15.3、2.5、2.8、12.3和1.2个百分点。改进后松枯死木识别算法具有较高精度,有效缓解了小目标与纹理模糊目标识别困难问题,为后续疫木清零提供技术支持。展开更多
基金the EBOR-project funded by the Academy of Finland (proj.no.276255)
文摘Background： After their death, Scots pine trees can remain standing for decades and sometimes up to 200 years,forming long-lasting and ecologically important structures in boreal forest landscapes. Standing dead pines decay very slowly and with time develop into ‘kelo＇ trees, which are characterized by hard wood with silvery-colored appearance. These kelo trees represent an ecologically important, long lasting and visually striking element of the structure of natural pine-dominated forests in boreal Fennoscandia that is nowadays virtually absent from managed forest landscapes.Methods： We examined and mapped the amount, structural features, site characteristics and spatial distribution of dead standing pine trees over a ten hectare area in an unmanaged boreal forest landscape in the Kalevala National Park in Russian Viena Karelia.Results： The mean basal area of dead standing pine trees in the forested part of the landscape was 1.7 m^2·ha^-1 and the estimated volume 12.7 m^3·ha^-1. From the total number of standing dead pine trees 65% were kelo trees, with a basal area of 1.1 m^2·ha^-1 and volume of 8.0 m^3·ha^-1, the remainder consisting of standing dead pines along the continuum between a recently dead tree and a kelo tree. Overall, standing dead pines were distributed throughout the study area, but there was a tendency towards spatial clustering up to 〈100 m distances. Standing dead pines were most commonly situated on flat ground or in the mid slope in the local topography.In addition, standing dead pines contributed to substrate diversity also by commonly having charred wood and broken tops. Based on the presence of dead pine snags in different stage of transition from a recently dead pine to a kelo with silvery surface, it seems evident that the process of kelo recruitment was continuously in action in the studied landscape.Conclusions： Kelo trees are an omnipresent feature in natural pine-dominated forest landscapes with important contribution to forest structural and substrate diversity. Because of their longevity and extremely slow turnover dynamics and importance for biodiversity, protection of vulnerable kelo tree populations, and ensuring their continuous recruitment, should be of high priority in forest restoration and sustainable management.
文摘为解决山地地形起伏大、无人机飞行高度高导致图像中尺度小且纹理模糊的松枯死木识别困难问题,该研究提出了一种在特征层级进行超分辨率重建的YOLOv5松枯死木识别算法。在YOLOv5网络中添加选择性核特征纹理迁移模块生成有细节纹理的高清检测特征图,自适应改变感受野的机制分配权重,将更多注意力集中在纹理细节,提升了小目标和模糊目标的识别精度。同时,使用前景背景平衡损失函数抑制背景噪声干扰,增加正样本的梯度贡献,改善正负样本分布不平衡问题。试验结果表明,改进后算法在交并比(intersection over union,IoU)阈值取0.5时的平均精度均值(mean average precision,mAP50)为92.7%,mAP50~95(以步长0.05从0.5到0.95间取IoU阈值下的平均mAP)为62.1%,APsmall(小目标平均精度值)为53.2%,相比于原算法mAP50提高了3.2个百分点,mAP50~95提升了8.3个百分点,APsmall提升了15.8个百分点。不同算法对比分析表明,该方法优于Faster R-CNN、YOLOv4、YOLOX、MT-YOLOv6,QueryDet、DDYOLOv5等深度学习算法,mAP50分别提高了16.7、15.3、2.5、2.8、12.3和1.2个百分点。改进后松枯死木识别算法具有较高精度,有效缓解了小目标与纹理模糊目标识别困难问题,为后续疫木清零提供技术支持。
