基于SFAM神经网络集成的土地评价  

Land evaluation based on SFAM neural network ensemble

作　　者：薛月菊[1] 胡月明[2] 杨敬锋[1] 陈强[1] 

机构地区：[1]华南农业大学南方农业机械与装备关键技术省部共建教育部重点实验室,广州510642 [2]华南农业大学信息学院,广州510642

出　　处：《农业工程学报》2008年第3期184-188,共5页Transactions of the Chinese Society of Agricultural Engineering

基　　金：广东省自然科学基金项目(04300504,05006623);广东省科技攻关项目(2005B20701008,2005B10101028,2004B20701006);国家科技攻关项目(2002BA516A08);国家星火计划项目(2006EA780057)

摘　　要：SFAM(Simplified Fuzzy ARTMAP,简化的模糊ARTMAP)神经网络具有自组织反馈、增量式学习和高度复杂映射等特点,是一种较BP神经网络和RBF神经网络等前馈神经网络更优秀的自组织神经网络。为克服SFAM神经网络受输入样本顺序的影响,提高土地评价的精度,提出利用SFAM神经网络集成进行土地评价的方法。并用SFAM神经网络、SFAM神经网络集成、BP神经网络、BP神经网络集成、RBF神经网络和RBF神经网络集成等方法对广东省中山市的土地进行了评价,对评价结果进行了分析和比较,结果表明SFAM神经网络具有比BP神经网络和RBF神经网络更优越的评价性能;对于这三种不同的神经网络,神经网络集成的土壤评价精度分别高于单个神经网络的精度。Simplified Fuzzy ARTMAP （SFAM） neural network is characterized as self-organized feedback, incremental learning and highly complex mapping, which outperforms forward neural networks such as BP （Backward Propagation） neural network and RBF （Radial Basis Function） neural network. To overcome the influence of the ordering of training sample presentation and improve the accuracy of land evaluation, a land evaluation method using SFAM neural network ensemble was presented. Moreover, SFAM neural network, SFAM neural network ensemble, BP neural network, BP neural network ensemble, RBF neural network, and RBF neural network ensemble were used to evaluate the land in Zhongshan city of Guangdong Province, China. And the results were analyzed and compared. The experimental results demonstrate that the accuracy of land evaluation using SFAM neural network is higher than those of BP neural network and RBF neural network, respectively; for the three types of neural networks, neural network ensembles perform better than their single neural networks, respectively.

关 键 词：土地评价 神经网络集成 SFAM神经网络 BP神经网络 RBF神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] S29[自动化与计算机技术—控制科学与工程]