| Summary: | In recent years, the novel coronavirus has had profound impacts on various countries and industries worldwide. This study aims to investigate the novel coronavirus first identified in 2019, characterized as a novel single-stranded RNA virus that easily infects the human respiratory system. Currently, the screening and diagnosis of the novel coronavirus primarily rely on nucleic acid testing, and the machine learning-based classification screening often employs traditional convolutional neural networks. However, this approach is not suitable for real-time rapid virus detection. In this dissertation, we meticulously compare three sets of preprocessing methods and seven feature selection and extraction methods to determine the optimal data preprocessing and feature extraction techniques. Simultaneously, based on the traditional Extreme Learning Machine (ELM), we propose a two-layer classification network combining the Extreme Learning Machine-Radial Basis Function (ELM-RBF) and Sparse Representation Classification (SRC). Compared to traditional ELM classification networks, our method exhibits superior performance. Ultimately, the comprehensive neural network classification system proposed in this study (employing SNV+second-order derivative preprocessing, Relief algorithm for feature extraction, and ELMRBF-SRC as the classifier) achieves impressive accuracy, sensitivity, and specificity, reaching 94.27%, 88.19%, and 97.64%, respectively. This study demonstrates the feasibility of using near-infrared spectroscopy based on throat swab sample extracts for the relatively straightforward system classification of COVID-19.
|
|---|