Serum species identification using mid-infrared and far-infrared spectroscopy combined with neural network algorithms

Serum is a widely used biological fluid containing rich biological information, commonly employed in clinical diagnosis and medical treatment. However, since human serum contains genetic information of races and involves national information security, accurate identification of species serum is required for customs import and export trade. In this study, we propose a species serum identification method combining mid-infrared and farinfrared spectroscopy with neural network algorithms. By collecting spectra of 147 serum samples from 8 species and conducting preliminary analysis on specific spectral ranges, differences between spectra of certain species were identified. Subsequently, a Convolutional Neural Network (CNN) model was constructed for species identification and classification, achieving recognition accuracy of 95.00 % for human and non-human serums (binary classification) and 8 species of serums (octal classification). Through correlation analysis, the corresponding relationships between species serums and their spectral characteristic regions, as well as relevant molecular groups and chemical bonds, were clarified. This reveals the underlying mechanism for species recognition based on serum spectra. Furthermore, by nesting multiple segmented progressive sub-models and optimizing the weight ratios of spectral features, the accuracy of species recognition was improved to 98.95 % while the computational complexity was greatly reduced. These results provide a reliable and efficient method for rapid differentiation of species serums, with important implications for the identification of other biological fluids.