RamanNet: a generalized neural network architecture for Raman spectrum analysis

Abstract Raman spectroscopy provides a vibrational profile of the molecules and thus can be used to uniquely identify different kinds materials. This sort molecule fingerprinting has led widespread application spectrum in various fields like medical diagnosis, forensics, mineralogy, bacteriology, virology, etc. Despite recent rise spectra data volume, there not been any significant effort developing generalized machine learning methods targeted toward analysis. We examine, experiment, evaluate existing conjecture that neither current sequential models nor traditional are satisfactorily sufficient analyze spectra. Both have their perks pitfalls; therefore, we attempt mix best both worlds propose novel network architecture RamanNet. RamanNet is immune invariance property convolutional neural networks (CNNs) at same time better than for inclusion sparse connectivity. achieved by incorporating shifted multi-layer perceptrons (MLP) earlier levels extract features across entire spectrum, which further refined triplet loss hidden layers. Our experiments on 4 public datasets demonstrate superior performance over much more complex state-of-the-art methods, thus, potential become de facto standard