最小差异采样的主动学习图像分类方法

Minimal difference sampling for active learning image classification

针对委员会成员模型投票不一致性的度量问题,提出了一种基于最小差异采样的主动学习图像分类方法。该方法首先基于标注样本集的重采样结果构建决策委员会,然后利用投票概率较高的2个类别的概率值的差异来度量未标注样本集每个样本的投票不一致性,选择概率差异最小的样本交由人工专家标注,如此迭代更新分类器。将新方法与EQB算法及nEQB算法在多个数据集上进行实验对比,实验结果表明所提方法能够有效提高分类的准确率。还对组成决策委员会的成员模型的数目设置进行了分析和讨论,结果表明在相同的成员模型数目时所提方法比nEQB算法更为有效。

Aiming at the problem of measuring the voting disagreement of committee, a minimal difference sampling method for image classification was proposed. It selects the sample with the minimal difference of two highest class probabilities voted by committee. The experimental results show that this method effectively enhances the classification accuracy compared with EQB and nEQB. Furthermore, the influence of the number of models in the decision-making committee was analyzed and discussed. The experimental results show that the proposed method always outperforms nEQB with the same number of models.