基于BP和RBF神经网络模型的草坪质量综合评价  

A comprehensive evaluation of turfgrass quality based on a BP and RBF neural network model

作　　者：肖波[1,2] 宋桂龙[1] 韩烈保[1,2] 包永霞[1] 李飞飞[1] 陈爱霞 

机构地区：[1]北京林业大学草坪研究所,北京100083 [2]长江大学园艺园林学院,湖北荆州434025 [3]荆州市高级技工学校,湖北荆州434025

出　　处：《草业学报》2012年第4期275-281,共7页Acta Prataculturae Sinica

基　　金：国家林业局"948"项目(2011-4-50);北京市重点学科建设项目资助

摘　　要：依据现有的草坪质量评价指标体系,于2010年调查了20个草地早熟禾品种成坪后的11项指标,包括草坪的密度、质地、颜色、均一性、绿期、抗病性、盖度、耐践踏性、成坪速度、草坪强度以及生物量。然后,运用神经网络原理及Matlab神经网络工具箱,以其中的15个草地早熟禾品种成坪后的11项指标的实地调查值作为网络输入,以专家打分作为网络输出,通过不断调整网络训练参数,使网络性能达到最优,构建了草坪质量综合评价的BP和RBF神经网络模型,并给出了BP和RBF神经网络模型的分析方法及其Matlab实现步骤。利用训练好了的网络模型,对其余的5个草地早熟禾品种的综合质量评价得分进行网络预测,结果表明,RBF神经网络的预测误差均小于2%,而BP神经网络的预测误差均大于5%,因此,基于RBF神经网络模型的草坪质量评价结果比BP神经网络更准确,可以用于草坪质量综合评价。与常规的加权法、层次分析法或模糊综合评判法评价草坪质量相比,基于RBF神经网络模型的草坪质量综合评价,在一定程度上减少了评价中主观因素的影响,简化了计算步骤,为草坪质量综合评价提供了一种全新的思路。Based on the recent turfgrass quality evaluation system,eleven indexes（density,texture,color,uniformity,green period,disease resistance,coverage,traffic tolerance,seedling establishment,turf strength and biomass） were used to select 20 Poa pratensis cultivars in 2010.The values of eleven indexes from 15 of the P.pratensis cultivars were selected as input data for the system using the principles of neural networks and the Matlab neural network toolbox.The output was expert graded data.Performance optimization was carried out by running the neural networks with different parameters and then models of the BP and RBF neural networks for evaluation of turfgrass quality were established.Methods of establishing neural network models and steps for Matlab are listed.Quality of the other 5 P.pratensis cultivars was evaluated using the trained neural network model.The predicted errors of the RBF neural network were less than 2% and the predicted errors of the BP neural network were more than 5% when they were applied as a comprehensive evaluation of turfgrass quality.Therefore the RBF neural network with smaller error odds was able to provide a more accurate evaluation of turfgrass quality than the BP neural network and it can be used to evaluate turfgrass quality.Compared with traditional methods,such as the weighting method,analytic hierarchy process,and fuzzy synthesis,the RBF neural networks accuracy reduces the influences of subjective factors and simplifies the calculating procedures.It provides a new idea for comprehensive evaluation of turfgrass quality.

关 键 词：草坪质量综合评价 BP神经网络 RBF神经网络 

分 类 号：S688.4[农业科学—观赏园艺]