基于栈式降噪自动编码器的气体识别

Gas recognition based on stacked denoising autoencoders

为克服手工设计特征的繁杂过程以及特征不通用性,提高气体识别准确率,提出一种基于深度学习的气体定性识别方法,自动提取自适应的气体数据特征。实验基于UCI机器学习气体数据集,分别对比基于2层深度神经网络结构-栈式降噪自动编码器以及浅层机器学习算法的气体定性识别效果。实验结果表明,基于深度学习算法自动提取特征的过程更简单、通用,提高了气体识别的准确率,改善了传统方法的复杂流程。

To simplify the complex process of designing features by hands,to make the features more general,and to improve the classification accuracy of gases,agas recognition method based deep learning was proposed that could extract self-adaption features.Two methods based UCI machine learning database were compared respectively in the experiments.One was a two-level structure of deep neural network-stacked denoising autoencoders,and the other was a kind of shallow machine learning algorithms used in gas qualiative recognition.Results show that extracting features automatically with deep learning is a simpler and more universal way in gas recognition.The method not only improves the classification accuracy of gases,but also reduces the complexity of the process using traditional shallow machine learning algorithms.