摘要
森林生物量的估算对于全球碳平衡和环境保护至关重要。通过遥感等手段提取与森林生物量相关的单波段特征、植被指数、纹理特征、地形因子等特征参数,特征数量往往较多,影响预测精度。该文提出了一种后向迭代的随机森林(RF-RFE)特征选择方法,即利用随机森林算法计算特征重要度,采用后向迭代的方法逐步简化特征参数。以内蒙古大兴安岭地区的激流河林场为研究区域,以实验区2012年"资源三号"遥感影像和森林资源3类调查的样地数据为数据源,使用RF-RFE算法进行特征选择分析。实验结果表明,在森林生物量遥感反演过程中的RF-RFE特征选择不但能降低时间复杂度,而且保证了特征选择的精度。
Forest biomass estimation is the key to maintain the global carbon balance and protect the environment. Some characteristic parameters associated with the forest biomass can be extracted from the remote sensing image, including the single-band information, vegetation indexes, texture features and terrain factors, which have been used to estimate the forest biomass accurately. However, the quantity of characteristic parameters is huge, and too many parameters have a negative influence on the precision of prediction. A novel parameter selection method that combines the Random Forest and Recursive Feature Elimination （RF-RFE） is proposed to reduce the number of biomass parameters, and enhance the estima- tion accuracy of forest biomass. This study adopted the ZY-3Satellite Image and the plot data of forest inventory in 2012, and used RF-RFE to select the parameters and estimate the forest biomass in Jiliuhe forest farm in Greater Khingan Mountain. The results show that the RF-RFE parameters selection method could significantly reduce the time complexity and obviously improve the precision of feature selection.
出处
《测绘科学》
CSCD
北大核心
2017年第5期100-105,共6页
Science of Surveying and Mapping
基金
国家"863计划"项目(2013AA122003)
国家质量监督检验检疫总局公益性行业科研专项(201410308)
