基于IFOA-LSSVM算法的机载LiDAR森林生物量估测

Forest Biomass Estimation by Airborne LiDAR Based on IFOA-LSSVM Algorithm

利用LiDAR数据的三维结构信息,提取样地级点云变量并进行优化,通过与获取的地面调查数据相结合,构建基于果蝇算法优化最小二乘支持向量机的生物量估测模型。利用反向学习初始化、三维搜索与自适应更新步长改进果蝇优化算法;将该算法优化最小二乘支持向量机LSSVM参数(σ,γ);建立基于IFOA-LSSVM的森林生物量估测模型。IFOA-LSSVM模型估测生物量的均方根误差值只有67.2195 t/ha。崖柏型、铁杉型、云杉型IFOA-LSSVM模型估测生物量的均方根误差值分别为55.2787 t/ha、63.6967 t/ha、36.0813 t/ha;估测值与实测值的相关系数平方为96.68%、93.71%、91.28%。基于IFOA-LSSVM模型的生物量估测误差和拟合程度均优于FOA-LSSVM。IFOA-LSSVM估测模型具有泛化能力强、收敛速度快、寻优精度高的特点。

This paper uses the three-dimensional structural information of LiDAR data to extract sample-level point cloud variables and optimize these variables,and combines them with the acquired ground survey data to construct an optimized least squares support vector machine biomass estimation model based on the fruit fly algorithm.First,this paper improves the fruit fly optimization algorithm by using backward learning initialization,three-dimensional search,and adaptive update step size.Then the least squares support vector machine is optimized by the approving algorithm LSSVM parameters(σ,γ).Finally,a forest biomass estimation model based on IFOA-LSSVM is established.The root-mean-square error of biomass estimated by the IFOA-LSSVM modelreaches to 67.2195 t/ha.The root-mean-square errors of the estimated biomass of thuja,hemlock,and spruce IFOA-LSSVM models are 55.2787 t/ha,63.6967 t/ha,36.0813 t/ha,respectively,and the values of square of correlation coefficient R2 between the estimated value and the measured value increase to 96.68%,93.71%,91.28%,respectively.The error and fitting degree of the biomass estimation based on the IFOA-LSSVM model are better than those of FOA-LSSVM.The IFOA-LSSVM estimation model has the characteristics of strong generalization ability,fast convergence speed and high optimization accuracy.