Abstract: In this paper，we studied the hyperspectral classification with missing attribute values adopting decision tree C4.5，Naive Bayes and NBTree on the Weka platform.As the large hyperspectral data have numerous wave bands and high information redundancy，firstly，we extracted the spectral feature parameters from the spectral curve and choosed the right wave bands of the absorption peaks as input vector.Experimental results indicate that the classification error of uranium-uraninite classification model based on NBTree is minimum；also it has the best classification accuracy compared with Naïve Bayes and J4.8. However，NBTree needs more training time than Naïve Bayes and J4.8. Finally，we analyzed the results of the three classification algorithms.

Key words: hyperspectral, classification, spectral absorption feature parameters, decision tree, Naive Bayes, Naive Bayes Tree

摘要： 在Weka平台上，采用决策树C4.5、朴素贝叶斯、朴素贝叶斯树三种算法进行了带缺失属性值的高光谱分类研究。针对高光谱波段数众多、信息冗余量大的特点，首先对光谱曲线进行光谱特征参数提取，然后再选择合适的吸收峰波段作为输入向量来进行分类。实验表明，由NBTree建立的铀黑-沥青铀矿分类模型的分类误差最小，分类精度最高，其次是Na?觙veBayes和J4.8，但从训练时间来看，NBTree则高于NB和J4.8。最后，对三种分类算法的分类结果进行了分析。

关键词: 高光谱, 分类, 光谱吸收特征参数, 决策树, 朴素贝叶斯, 朴素贝叶斯树