3种模型在GF-2影像的生物量估测中的比较

Comparison of three models in biomass estimation of GF-2 images

为了研究高分二号(GF-2)影像生物量估测的模型效果,以攸县黄丰桥林场为研究区,在研究区内采用随机抽样的方法,结合国家森林资源连续清查样地,获取了共47个样地的生物量数据。对GF-2影像进行预处理,结合相关研究,提取8个单波段信息、24个多波段组合信息、4个植被指数以及海拔、坡度、坡向等39个因子作为建模的自变量,采用主成分分析、偏最小二乘和BP神经网络3种方法建立生物量估测模型。结果表明:主成分回归模型的实测值和预测值的决定系数R^2为0.44,模型的估测精度为65.83%;偏最小二乘回归模型的R^2为0.50,模型的估测精度为67.66%;BP神经网络模型的R^2为0.79,模型的估测精度为78.62%。比较可知,BP神经网络模型效果最好。

In order to study the model effect of biomass estimation on GF-2 images,this study takes Huangfengqiao forest farm in Youxian County as the research area,in the study area,a random sampling method was used to plot plots,and the biomass data of 47 plots were obtained by combining the sample plots of national forest resources.The GF-2 image pre-processing,combined with relevant research,extracting 8 single band information,24 multi band information,4 vegetation index,altitude,slope,and 39 factors such as modeling variables,using principal component analysis,partial least squares and BP neural network method to establish three kinds of biomass estimation model.The actual decision principal component regression model and predicted values of coefficient R^2 is 0.44,the estimation accuracy of the 65.83% models;the R^2 of the partial least squares regression model is 0.50,the estimation accuracy of the model is 67.66%;the R^2 of BP neural network model is 0.79,the estimation accuracy of the model is 78.62%,By comparison,BP neural network model is the best.