An electronic nose (E-nose) system is regularly composed of a gas sensor array and certain pattern-recognition algorithms. With the use of E-nose, the gas sensors inevitably undergo physical changes, which causes gas-sensor drift to invalid algorithm models of E-noses. In this study, we intend to explore a suitable approach for online E-nose drift calibration. Considering drift calibration samples cannot be obtained directly during continuous odor detection, we have adopted Active Learning (AL) paradigm to select calibration samples from previous tested samples and provide their categories by querying. Further, we deal with the class imbalance problem of drift calibration set caused by traditional AL instance-selection strategy. We propose a new strategy named Dual-Rule Sampling (DRS) to simultaneously measure sample uncertainty and minority-class similarity. The high uncertain instances being close to minority-class are selected for drift calibration when class imbalance occurs. We have used two datasets to evaluate the performance of DRS. The experimental results show that DRS reaches the highest recognition score among all the tested methodologies by emphasizing the minority-class recognition improvement. We can conclude that DRS successfully implements online E-nose drift calibration in continuous odor detection.

电子鼻（E-nose）系统通常由气体传感器阵列和某些模式识别算法组成。使用电子鼻，气体传感器不可避免地会发生物理变化，这会导致气体传感器漂移到无效的电子鼻算法模型。在这项研究中，我们打算探索一种适合的在线电子鼻漂移校准方法。考虑到在连续气味检测过程中无法直接获得漂移校准样品，我们采用主动学习（AL）范式从以前测试的样品中选择校准样品，并通过查询提供其类别。此外，我们处理了由传统的AL实例选择策略引起的漂移校准集的类不平衡问题。我们提出了一种称为双规则采样（DRS）的新策略，可以同时测量样本不确定性和少数族裔相似度。当类不平衡发生时，选择接近少数类的高不确定性实例进行漂移校准。我们使用了两个数据集来评估DRS的性能。实验结果表明，DRS通过强调少数族裔类别的识别改进，在所有测试方法中均达到最高的识别分数。我们可以得出结论，DRS在连续气味检测中成功实现了在线电子鼻漂移校准。实验结果表明，DRS通过强调少数族裔类别的识别改进，在所有测试方法中均达到最高的识别分数。我们可以得出结论，DRS在连续气味检测中成功实现了在线电子鼻漂移校准。实验结果表明，DRS通过强调少数族裔类别的识别改进，在所有测试方法中均达到最高的识别分数。我们可以得出结论，DRS在连续气味检测中成功实现了在线电子鼻漂移校准。