子空间分布自适应的传感器在线漂移补偿算法

Online Sensor Drift Compensation Algorithm Based on Subspace Distribution Adaptation

传感器漂移补偿方法针对电子鼻系统中气体传感器的输出响应随使用时间延伸发生改变,进而导致气体识别准确率下降这一问题提出,大多针对离线场景,在实际应用中存在定期进行系统人工校正所造成的成本高耗时长等困难。因此,本文提出了一种子空间分布自适应的传感器在线漂移补偿算法。算法通过构造测地线核将原始样本与漂移样本嵌入到流形子空间,然后引入条件分布自适应和流形正则化以减小样本特征的分布差异,并利用结构风险最小化原则构建分类器。分类模型的在线更新通过将每轮分类后获得预测标签的漂移样本引入到下一轮的模型训练过程中以实现。在公开数据集上进行漂移补偿实验,结果表明,提出的算法提高了漂移样本的分类准确率,有效地实现了传感器的在线漂移补偿。

Sensor drift compensation algorithm is aimed at the problem that output response of gas sensors in electronic nose system changes with the use of time,which leads to a decrease in accuracy of gas identification.Most of those algorithms are offline and not suitable for practical applications because of the high cost and time consuming of manual system calibrate periodically.For this problem,an online drift compensation algorithm based on subspace distribution adaptation(ODCSDA)is proposed.In ODCSDA,original samples and drift samples are embedded into manifold subspace by constructing a geodesic flow kernel firstly.Then conditional distribution adaptation and manifold regularization is introduced to reduce the difference of feature distributions between samples.Finally,a classifier is constructed by structural risk minimization principle.The online update of the classification model is implemented by introducing the drift samples which have obtained predicted labels after each round of classification into the next round of model training.ODCSDA is used to perform drift compensation experiments on a public dataset.Results show that ODCSDA can improve the classification accuracy of drift data and realize online sensor drift compensation effectively.