摘要
首先介绍了CART、C5.0和概率神经网络三种机器学习算法的原理,然后以覆盖湖北省公安县的ALOS影像为数据源,从整体精度、对训练样本大小和噪声的敏感性三个方面对它们进行了比较分析。结果显示C5.0算法分类的整体精度最高,达到83.59%。概率神经网络受训练样本大小和噪声的影响最低:在训练样本大小降为原样本数据量的40%时,其精度为78.52%;噪声占训练样本量的10%时,精度只下降了4.3%。通过分析可以看出,在训练样本量充足时,C5.0算法的分类精度最好,而在样本不足或者包含噪声的情况下,使用概率神经网络算法能比其他两种算法取得更好的分类效果。
This paper firstly introduces the principles of three machine learning algorithms which are CART,C5.0 and Probabilistic Neural Networks,and then makes a comparative analysis from overall accuracy,size of data set and noise sensitivity,using the ALOS image of Gongan County,Hubei Province as the study data source.The result indicates that the overall accuracy of C5.0 algorithm is the highest,up to 83.59%.The affection from the data set size and the noise on Probabilistic Neural Networks is the lowest：the accuracy is 78.52% when the training set size reduces to 40% of the original size;while the noise accounts for 10% of the original set size,the accuracy only decreases by 4.3%.From the analysis it can be seen that the overall accuracy of C5.0 algorithm is the highest when the training set size is sufficient;however,in case of sample insufficient or containing noise,it can achieve better classification effect than the other two by using Probabilistic Neural Network algorithm.
出处
《遥感信息》
CSCD
2010年第3期26-29,111,共5页
Remote Sensing Information
基金
国家自然科学基金重点项目(40730635)
水利部公益项目(200701024)
关键词
机器学习算法
ALOS影像
分类
土地覆盖
machine learning algorithms
ALOS image
classification
land cover
