基于不确定抽样的半监督城土地功能分类方法-吉林大学学报.PDFVIP

第34卷摇 第4期 吉林 大学学报 ( 息科 学版) Vol.34摇 No.4 2016年7月 Journal of Jilin University (Information Science Edition) July2016 文章编号:1671鄄5896(2016)04鄄0550鄄06 基于不确定抽样的半监督城市土地功能分类方法 蔡摇 柳,恵摇 飞,叶摇 敏,康摇 科,赵祥模 (长安大学 信息工程学院,西安710064) 摘要:为提高分类准确率,解决城市区域社会功能标签分类难的问题,提出了一种基于不确定抽样选择策略的 半监督城市土地功能分类方法。 该算法从轨迹数据中提取城市区域的特征向量,只需对少量区域进行标签,根 据不确定抽样的主动学习选择策略,从未标注训练样本中选取具有较多信息的数据,利用半监督学习算法进行 标注,得到新的标注训练样本添加到训练集,反复迭代后得到分类结果。 实验结果表明,该方法对不同社会功 能的城市区域分类准确率可达90.2%,与传统方法相比分类准确率高,减少了大量标注工作,在少数标签数据 上仍有较好的分类效果。 关键词:轨迹数据挖掘;半监督学习;不确定抽样;土地功能分类 中图分类号:TP391.4 文献标识码:A Semi鄄Supervised Urban Land Using Classification Method Based on Uncertainty Sampling CAI Liu,HUI Fei,YE Min,KANG Ke,ZHAO Xiangmo (College of Information Engineering,Chang蒺an University,Xi蒺an710064,China) Abstract:Inorderto improvetheclassification accuracy and solvetheproblemof thedifficulty of labeling social functionsof theurbanland,a semi鄄supervisedurbanlandfunction classificationmethodbasedontheuncertainty sampling selection strategy is proposed. The algorithm extracts the feature vector of the urban area from the trajectory data,and only a small number of areas need to be labeled. According to the uncertainty sampling active learning selection strategy,unlabeledtraining sampleswith more information data are selected andlabeled by semi鄄supervised learning algorithm. The new labeled training samples are added to the training set. The classification resultswill come out after repeated iteration. Experimental results show that the proposed method can reach the accuracy rate of 90.2% on classifying urban areas with different social functions. It has a high classification accuracy and reduces a lot of labeling work compared to traditional methods, showing a good