论文部分内容阅读
为了解决局部支持向量机算法KNNSVM存在的分类时间过长不利于具有海量数据量的高分辨率遥感图像分类的不足,提高KNNSVM的算法表现,提出了改进的基于不确定性的BKNNSVM算法.该算法利用二项式分布的共轭先验分布Beta分布根据近邻的分布情况推导该未标记样本属于正类或负类的概率大小,从而计算每一个未标记样本在类属性上的不确定性大小.再通过设置不确定性阈值的大小,对不确定性低于阈值的未标记样本直接采用KNN进行分类,而对高于阈值的样本利用其近邻建立局部支持向量机分类器进行分类.对高分辨率图像分类的实验结果表明:合适的阈值能够有效降低原始KNNSVM算法的时间开销,同时能保持KNNSVM分类精度高的特点.
In order to solve the shortcoming of the local support vector machine algorithm KNNSVM that the classification time is not conducive to the classification of high-resolution remote sensing image with massive data amount and to improve the performance of KNNSVM, an improved BKNNSVM algorithm based on uncertainty is proposed. Conjugate primal distribution Beta distribution using binomial distribution The probability of the unlabeled sample belonging to the positive or negative class is deduced according to the distribution of neighbors, so as to calculate the uncertainty of class size of each unlabeled sample. Then, by setting the size of the uncertainty threshold, KNN is used to classify the unlabeled samples whose uncertainty is below the threshold and the local SVM classifiers are used to classify the samples above the threshold. The experimental results of rate image classification show that the suitable threshold can effectively reduce the time overhead of the original KNNSVM algorithm and maintain the high classification accuracy of KNNSVM.