基于特征提取与选择的气体识别研究

Research on Gas Recognition Based on Feature Extraction and Selection

在电子鼻系统中,特征提取和选择以及分类模型都是其性能改进的关键。针对从传感器阵列中提取单一特征时会忽略传感器特异性的问题,提出基于相关性分析来选择每一个传感器最优的特征提取方法,组成最优特征向量进行气体识别,实验表明:通过该方式提取的特征向量在分类模型中表现更好,在各模型的平均识别准确率提升了0.027,其中支持向量机和人工神经网络提升效果最明显,分别提升了0.031和0.054。并根据模型特性和实际需求,提出逻辑回归与支持向量机结合的二次分类模型,实验表明该模型能够进一步提高分类准确率,降低具体气体检测场景中辨别气体错误的风险。

In the electronic nose system,both feature extraction and selection as well as classification models are the keys to its performance improvement.Aiming at the problem of ignoring the sensor specificity when extracting a single feature from the sensor array,this paper proposed to select the optimal feature extraction method for each sensor based on correlation analysis,and form the optimal feature vector for gas identification.The experiment results show the eigenvectors perform better in the classification model,and the average recognition accuracy of each model increases by 0.027,of which the support vector machine and artificial neural network have the most obvious improvement,with an increase of 0.031 and 0.054,respectively.According to the model characteristics and actual requirements,a secondary classification model combining logistic regression and support vector machine is proposed.Experiments show that the model can further improve the classification accuracy and reduce the risk of gas identification errors in specific gas detection scenarios.