决策树学习方法应用于生境景观分类

Decision tree learning for habitat landscape classification

遥感影像分类是进行景观生态学研究以及区域景观规划的基础工作,是获取环境资源与信息的主要手段。研究选取秦岭南坡地区100km2范围为实验区域,综合遥感影像、数字高程模型等空间数据,利用C5.0决策树学习算法从750个实地调查样点中自动提取分类知识、建立规则库并实现计算机自动景观分类;同时分析根据不同数量样点得到的决策树规则以及决策树分类精度变化的趋势。分析结果表明,在样点信息充分的条件下,利用决策树学习方法能够实现高景观分类精度;随着样点数量的增加,分类精度也随之提高,该研究中景观分类精度最高达到79.0%。

A landscape classification knowledge was mined from ground-truthing data using decision tree learning for knowledge-based habitat landscape classification. 750 field samples were collected from a remote sensing and digital elevation data set within a 100-km^2 research area on the southern slope of the Qinling Mountains. The expert classification rules were defined using the C5.0 algorithm for the knowledge-based classification. The influence of different numbers of sample points on the knowledge accuracy was then evaluated. The results show that the knowledge can be conveniently mined using decision tree learning with a sufficient number of sample points. The classification accuracy increases as the number of sample points increases. The decision tree learning classification gave a highest accuracy of 79.0%.