基于对抗学习的电子鼻传感器漂移补偿算法

Electronic nose sensor drift compensation algorithm based on adversarial learning

电子鼻是一种生物嗅觉系统,主要由气体传感器阵列和模式识别组成,并已应用在生活的许多领域。但是在电子鼻实际应用中传感器易发生漂移,致使电子鼻性能下降。本文提出一种基于对抗学习估计域不变原型方法用于补偿传感器漂移。该算法包含一个由神经网络构成的特征提取器和分类器,并利用条件熵度量无标记目标域特征和估计原型(每一类的表示)的相似度。为使目标域特征更具有区分性,训练分类器最大化熵,训练特征提取器最小化熵。最后,实验结果表明:该算法能够有效减少电子鼻传感器漂移。

Electronic nose(E-nose)is a bionic olfactory system,which is mainly composed of gas sensor array and pattern recognition,and has been applied to many fileds in our life.However,sensor drift is easy to occur in realistic application scenario of E-nose,which makes a decrease in performance of E-nose.Aiming at this problem,a method,namely estimate domain-invariant pototypes via adversarial learning(ALDIP),is put forward for sensor drift compensation.The basic model for the algorithm includes a feature extractor and classifier composed of neural networks,and uses the conditional entropy to calculate the similarity between the unlabeled target domain features and the estimated prototypes(representatives of each class).In order to make features of the target domain have more discrimination,train the classifier to maximize entropy and minimizes it with respect to the feature extractor.Finally,experiments show that the algorithm can effectively reduce drift of E-nose sensor.