eduzhai > Applied Sciences > Engineering >

Feature visualization of Raman spectrum analysis with deep convolutional neural network

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: We demonstrate a recognition and feature visualization method that uses adeep convolutional neural network for Raman spectrum analysis. Thevisualization is achieved by calculating important regions in the spectra fromweights in pooling and fully-connected layers. The method is first examined forsimple Lorentzian spectra, then applied to the spectra of pharmaceuticalcompounds and numerically mixed amino acids. We investigate the effects of thesize and number of convolution filters on the extracted regions for Raman-peaksignals using the Lorentzian spectra. It is confirmed that the Raman peakcontributes to the recognition by visualizing the extracted features. Anear-zero weight value is obtained at the background level region, whichappears to be used for baseline correction. Common component extraction isconfirmed by an evaluation of numerically mixed amino acid spectra. High weightvalues at the common peaks and negative values at the distinctive peaks appear,even though the model is given one-hot vectors as the training labels (withouta mix ratio). This proposed method is potentially suitable for applicationssuch as the validation of trained models, ensuring the reliability of commoncomponent extraction from compound samples for spectral analysis.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×