基于CGAN增强的新冠病毒拉曼光谱深度学习分类研究

Deep Leaning Classification of Novel Coronavirus Raman Spectra Enhanced by CGAN

新型冠状病毒的出现给全球经济、公共安全等方面带来严重损失。如今新型冠状病毒仍然在全球肆虐,加强病毒的检测和诊断是一项急需解决的工作,快速且准确地诊断新冠病毒对疫情防控尤为重要。随着技术的发展,深度学习在检测识别方面有了许多突破性进展,吸引着研究者的广泛关注。然而深度学习需要大量数据进行模型训练,而新冠病毒的拉曼光谱采集过程受到器件和环境等诸多问题的限制,导致获取大批量的数据非常困难。有限的训练数据会阻碍深度学习模型的训练,导致深度学习模型训练受阻,精度有限,以至于深度学习模型在真正的检测中表现不佳。为解决这一问题,该研究引入条件生成对抗网络,自动提取拉曼光谱数据特征,通过学习生成新的光谱来完成新冠病毒的拉曼光谱数据扩充工作。这些方法能有效增加训练集数量从而提高模型准确性。利用数据增强后的拉曼光谱数据和深度神经网络以及传统的机器学习方法逻辑回归、决策树、随机森林、支持向量机、KNN算法,完成新冠病毒的诊断。实验结果表明,深度神经网络对是否感染新型冠状病毒的预测准确率达到98%,高于传统机器学习算法,证明了深度学习模型在新冠病毒拉曼光谱检测中的优越性。该研究还对比了增强前后的数据集在不同模型中的表现,证明了数据增强的作用。与传统的检测方法相比,该方法具有无损、快速、准确等特点,为生物医学检测新冠病毒提供了一种新的思路。该方法可以为快速准确检测新冠病毒提供辅助,所采用的方法不仅可以应用于新冠病毒检测,还可以推广到其他疾病的诊断中,具有一定的实际应用价值。

The emergence of the novel coronavirus has caused significant losses in the global economy and public safety.The need for efficient detection and diagnosis has become urgent as the virus continues to inflict damage worldwide.Accurate and fast diagnosis of the novel coronavirus is critical for epidemic prevention and control.With the development of technology,deep learning has made many breakthroughs in detection and recognition,attracting widespread attention from researchers.However,deep learning requires a large amount of data for model training,and the collection of Raman spectra data for the novel coronavirus is limited by devices and environmental factors,making it difficult to obtain large amounts of data.The limited training data can hinder the training of deep learning models and limit their accuracy,resulting in poor performance in actual detection.To address this problem,this study introduces the CGAN adversarial network to automatically extract features from Raman spectra data and generate new spectra to expand the dataset for the novel coronavirus.These methods can effectively increase the training set's size and improve the model's accuracy.Using the enhanced Raman spectra dataset and deep neural networks,as well as traditional machine learning methods,including logistic regression,decision trees,random forests,support vector machines,and k-nearest neighbors algorithm,the diagnosis of COVID-19 is performed.The experimental results show that the deep neural network has a prediction accuracy of 98%for whether a patient is infected with the novel coronavirus,which is higher than traditional machine learning algorithms,demonstrating the superiority of deep learning models in Raman spectra detection of the novel coronavirus.We also compared the performance of the datasets before and after augmentation in different models,proving the effectiveness of data augmentation.Compared with traditional detection methods,this method is non-invasive,fast,and accurate,providing a new approach for biomedical detection of the novel coronavirus.The method proposed in this study can assist in rapidly and accurately detecting the novel coronavirus.It can be applied not only to the diagnosis of the novel coronavirus but also to the diagnosis of other diseases,having practical application value.