文献类型：期刊文献

中文题名：基于自适应神经模糊系统的杉木冠幅估算方法

英文题名：A Method of Estimating Chinese Fir Crown Width Based on Adaptive Neuro-Fuzzy Inference System

作者：李永亮[1] 张怀清[1] 杨廷栋[1] 马载阳[1] 李思佳[1] 沈康[1]

第一作者：李永亮

通信作者：Zhang, Huaiqing

机构：[1]中国林业科学研究院资源信息研究所

年份：2019

卷号：55

期号：11

起止页码：45-51

中文期刊名：林业科学

外文期刊名：Scientia Silvae Sinicae

收录：CSTPCD;;EI(收录号：20200808196070);Scopus;北大核心:【北大核心2017】；CSCD:【CSCD2019_2020】；

基金：中央级科研院所基本科研业务费专项（CAFYBB2017SZ005）

语种：中文

中文关键词：自适应神经模糊系统;冠幅;估算;空间结构单元;智能化

外文关键词：adaptive neuro-fuzzy inference system;crown width;estimation;spatial structural unit;intelligentialize

分类号：S757

摘要：【目的】基于相邻木特征与对象木冠幅间的复杂关系,提出一种基于自适应神经模糊系统的冠幅估算方法,以提高林木冠幅智能化估算水平。【方法】以杉木为研究对象,根据相邻木相对对象木的距离和方位,采用象限补树法构建空间结构单元。测定100组4方向冠幅、距离和方位角,提出相邻木冠幅、距对象木距离2个自变量的计算方法,以对象木冠幅与相邻木冠幅的比值作为因变量。根据样本数据,分析变量间非线性映射关系,建立25条模糊逻辑推理规则,设计以2个自变量为输入、1个因变量为输出的零阶Takagi-Sugeno模型,以70组数据训练自适应神经模糊系统,以30组数据检验系统冠幅估算效果,并与多元线性回归法和BP神经网络法进行对比。【结果】3种方法冠幅估算值与真实值的线性关系均达显著水平,经检验,本研究方法、BP神经网络法和多元线性回归法的判定系数分别为0.71、0.67和0.66。【结论】基于自适应神经模糊系统的冠幅估算方法可在自变量不含对象木属性特征的情况下,根据空间结构单元内相邻木特征,直接实现对象木冠幅的智能化估算。
【Objective】 Aiming at the complex relationships between neighborhood trees characteristics and subject tree crown width, a method of estimating crown width based on adaptive neuro-fuzzy inference system was proposed to improve the intelligent level of estimating crown width.【Method】 Chinese fir was taken as the research object. According to the distance and azimuth of the neighborhood trees relative to the subject tree, a method of adding trees in quadrants was presented to build spatial structural units. One hundred sets of data which contained crown width in four aspects, distances and azimuths were measured, computing method of two independent variables, including neighborhood crown width and distance from the subject tree, were proposed, and then the ratio of the subject tree crown width to the neighborhood crown width was defined as the dependent variable. According to the sample data, the nonlinear mapping relations among variables were analyzed, and twenty-five fuzzy logic inference rules were established. Azero-order Takagi-Sugeno model, which was composed of two inputs and one output, was designed. The adaptive neuro-fuzzy inference system was trained by seventy sets of data, and tested by thirty sets of data. It was contrasted with multi-element linear regression and back propagation neural network.【Result】 The linear relationships of crown width estimated by three method and the true value all reached significant levels. On inspection, determination coefficients of this method, back propagation neural network and multi-element linear regression were 0.71, 0.67, and 0.66, respectively.【Conclusion】 According to the neighborhood trees characteristics in the spatial structural unit, this method could directly, effectively and intelligently estimate the subject tree crown width without the independent variables those contain attributes of the subject tree.