Rank order ,ernels for the classification of Raman spectra of bacteria

Abstract

Bacterial identification is one of the applications for which classification using Raman spectra has proved to be successful. In this paper, we propose the use of Rank Order Kernels to classify Raman spectra in order to identify bacterial samples. Rank Order Kernels are two-dimensional image functions. The first step in the process transforms each Raman spectrum to a two-dimensional image. This is achieved by splitting the spectra into segments and calculating the ratio between the mean value of each and every other segment. The resulting two-dimensional matrix of ratios for each Raman spectrum is the image processed by the Rank Order Kernels. A similarity metric is used with a nearest neighbor algorithm for classification. The metric is based on rank order kernels. Our results show that the rank order kernel method is comparable in accuracy to other previously-used methods.