基于GAN-CNN的气味识别

Odor Identification Based on GAN-CNN

随着嗅觉技术的发展,电子鼻因其响应速度快、使用方便等特点被广泛的应用。然而,在实践中,很难获得足够的样本数据来训练气味识别模型。本文提出了一种将生成对抗网络(GAN)和卷积神经网络(CNN)相结合的模型GAN-CNN,首先,将电子鼻数据转换为灰度图。然后,利用GAN学习真实样本的数据分布,通过生成器和判别器的对抗训练,生成具有相似数据分布的增强数据集。最后,将增强数据集输入CNN进行特征提取,并对气味类型进行分类。结果表明,GAN-CNN利用电子鼻采集响应成功对气味进行分类,并且准确度达到94.43%,这对于小样本的气味识别具有重要意义。

With the development of olfactory technology,electronic noses are widely used due to their fast response speed and conve⁃nient use.However,in practice,it is difficult to obtain enough sample data to train the odor recognition model.This paper proposes a model GAN-CNN that combines Generative Adversarial Networks(GAN)and Convolutional Neural Networks(CNN).First,the elec⁃tronic nose data is converted into grayscale images.Then,GAN is used to learn the data distribution of real samples,and through the confrontation training of the generator and the discriminator,an enhanced data set with similar data distribution is generated.Finally,input the enhanced data set into CNN for feature extraction and classify the odor type.The results show that GAN-CNN successfully classifies odors by using the electronic nose to collect responses,and the accuracy reaches 94.43%,which is of great significance for odor recognition of small samples.