广东省典型农作物智能专家分类系统设计与示范应用  

The Design and Implementation of Intelligent Expert Classifier for Typical Crop Mapping

作　　者：方立刚[1,2] 李宏丽[1] 陈水森 

机构地区：[1]苏州市职业大学计算机工程系,江苏苏州215104 [2]广东省地理空间信息技术与应用公共实验室,广州510070

出　　处：《农机化研究》2011年第5期35-38,共4页Journal of Agricultural Mechanization Research

基　　金：农业部资源遥感与数字农业重点实验室开放基金项目(RDA0862);广东省科技计划重点引导项目(2006B1001014)

摘　　要：在广东省科技计划项目支持下,在广东省典型农作物标准波谱数据库中设计了符合典型农作物的长势监测和种植面积测算功能模块,将数据挖掘和知识发现应用到专家分类算法中,自行开发了适合农作物数据发现和挖掘的归纳学习算法,充分利用了波谱库中大量的波谱数据、相关属性和空间数据,形成了基于波谱库的广东省典型农作物智能专家分类系统。同时,进行了深圳市的荔枝面积估算的示范研究,结果显示分类精度达到91.8%,KAPPA达到0.81,高于传统分类算法,能满足农业部门对农情监测的要求。With the support of the science and technology project of Guangdong Province,the research design a module which is used to monitor crop condition and estimate area in spectral library of typical crops of South China and here present an approach to combine inductive learning with expert system classification methods.Spatial data mining techniques are used to discover knowledge from spectral library for expert system and inductive learning algorithm,which is designed specially for typical crops data,is developed independently.It makes use of a lot of spectral data,attribute data and spatial data of spectral library,and form intelligent expert classifier for South China＇s typical crops based on spectral library.The estimation of lychee planting area of Shenzhen city in 2007 is presented as a case study.Compared with traditional unsupervised classification,it improves classification accuracy greatly,and classification accuracy is reached at 91.8% with KAPPA coefficient of 0.81.The result indicates that intelligent expert classifier is a good classification method for lychee planting area extraction and be able to content with the need in agriculture application for quick crop area monitoring.But the classification algorithm has still some defects.First,our inductive learning algorithm is comparatively simple,only based on pixel granularity to select rule.At present,there exist some difficulties for the knowledge discovery of complicated temporal-spatial data.Secondly,the image classification is restricted by only assigning a kind of land cover category to a pixel.

关 键 词：波谱库 典型农作物 知识挖掘 专家分类 归纳学习 

分 类 号：TP182[自动化与计算机技术—控制理论与控制工程]