BioTwinNet: Dual-Stream Multilevel Feature Fusion for Classification of SARS-CoV-2 and Influenza Virus Variants via Genomic Image Processing

The fast evolution and high transmissibility of emerging infectious illnesses like Influenza and SARS-CoV-2 pose serious public health problems. In order to effectively classify SARS-CoV-2 and Influenza variations, this study presents BioTwinNet, a unique dual-stream deep learning framework that integrates multilevel feature fusion with genomic image processing. Using methods like Wavelet Analysis, Discrete Cosine Transform, and Power Spectral Density, DNA sequences were converted into visual representations. BioTwinNet, which makes use of the depth capabilities of ResNet architectures and the hierarchical feature extraction of Convolutional Neural Networks, was used to evaluate these genomic images. By using Discrete Cosine Transform-Power Spectral Density characteristics, this method achieves an impressive classification accuracy of 97.42%, overcoming the difficulties caused by differences in sequence length.