在三维重建中,运动恢复结构法(structure from motion,Sf M)作为经典重建方法,离不开图像的特征提取与初始图像对的选取。在利用标准ResNet-18算法进行图像特征提取时会遇到特征提取速度较慢、参数数量过多等问题。为此,利用改进的ResNe...在三维重建中,运动恢复结构法(structure from motion,Sf M)作为经典重建方法,离不开图像的特征提取与初始图像对的选取。在利用标准ResNet-18算法进行图像特征提取时会遇到特征提取速度较慢、参数数量过多等问题。为此,利用改进的ResNet-18处理部分上游任务,通过将标准残差块替换为深度可分离卷积,可以降低网络参数数量,提高图像特征提取与匹配方面的效率。另外,将ReLU激活函数更换为ELU激活函数,可以提高特征提取质量,进一步优化运动恢复结构法的相关性能。实验结果表明,优化后的算法在两个测试集上的图像特征提取耗时减少了48.45%,网络参数数量也得到了一定的降低,提高了运算效率,同时也在一定程度上优化了网络结构,得到了良好的提取效果。展开更多
Biometric inventories for 25 years,from 1983 to 2005,indicated that the Jianfengling tropical mountain rain forest in Hainan,China,was either a source or a modest sink of carbon.Overall,this forest was a small carbon ...Biometric inventories for 25 years,from 1983 to 2005,indicated that the Jianfengling tropical mountain rain forest in Hainan,China,was either a source or a modest sink of carbon.Overall,this forest was a small carbon sink with an accumulation rate of(0.56±0.22) Mg C ha-1yr-1,integrated from the long-term measurement data of two plots(P9201 and P8302).These findings were similar to those for African and American rain forests((0.62±0.23) Mg C ha-1yr-1).The carbon density varied between(201.43±29.38) Mg C ha-1 and(229.16±39.2) Mg C ha-1,and averaged(214.17±32.42) Mg C ha-1 for plot P9201.Plot P8302,however,varied between(223.95±45.92) Mg C ha-1 and(254.85±48.86) Mg C ha-1,and averaged(243.35±47.64) Mg C ha-1.Quadratic relationships were found between the strength of carbon sequestration and heavy rainstorms and dry months.Precipitation and evapotranspiration are two major factors controlling carbon sequestration in the tropical mountain rain forest.展开更多
文摘在三维重建中,运动恢复结构法(structure from motion,Sf M)作为经典重建方法,离不开图像的特征提取与初始图像对的选取。在利用标准ResNet-18算法进行图像特征提取时会遇到特征提取速度较慢、参数数量过多等问题。为此,利用改进的ResNet-18处理部分上游任务,通过将标准残差块替换为深度可分离卷积,可以降低网络参数数量,提高图像特征提取与匹配方面的效率。另外,将ReLU激活函数更换为ELU激活函数,可以提高特征提取质量,进一步优化运动恢复结构法的相关性能。实验结果表明,优化后的算法在两个测试集上的图像特征提取耗时减少了48.45%,网络参数数量也得到了一定的降低,提高了运算效率,同时也在一定程度上优化了网络结构,得到了良好的提取效果。
基金supported by the National Natural Science Foundation of China (Grant No. 40601009),the A3 Foresight Programthe Special Research Program of Research Institute for Tropical Forestry,Chinese Academy of Forestry (Grant No. 2007-1, RITFYWZX200902)+2 种基金 the Special Research Program for Public-welfare Forestry (Grant No. 200804001) the Eleventh-Five-Year Scientific Support Program of the State Forestry Administration of China (Grant Nos. 2006BAD03A07 and 2006BAD03A04)the "948" Program of State Forestry Administration of China (Grant No. 2007-4-18)
文摘Biometric inventories for 25 years,from 1983 to 2005,indicated that the Jianfengling tropical mountain rain forest in Hainan,China,was either a source or a modest sink of carbon.Overall,this forest was a small carbon sink with an accumulation rate of(0.56±0.22) Mg C ha-1yr-1,integrated from the long-term measurement data of two plots(P9201 and P8302).These findings were similar to those for African and American rain forests((0.62±0.23) Mg C ha-1yr-1).The carbon density varied between(201.43±29.38) Mg C ha-1 and(229.16±39.2) Mg C ha-1,and averaged(214.17±32.42) Mg C ha-1 for plot P9201.Plot P8302,however,varied between(223.95±45.92) Mg C ha-1 and(254.85±48.86) Mg C ha-1,and averaged(243.35±47.64) Mg C ha-1.Quadratic relationships were found between the strength of carbon sequestration and heavy rainstorms and dry months.Precipitation and evapotranspiration are two major factors controlling carbon sequestration in the tropical mountain rain forest.