A New Method for Detecting known components in Multi-level Raman Spectra
A new approach, called window adjusted lack of fit (WALOF), is introduced to improve the original LOF-based classification method. Basically, we introduce a range of additional measurements to further investigate the results close to the criterion boundaries by the original LOF-based classification. These additional measurements are carried out in terms of the peak regions in the observation spectra and reference components.
LOF method is quick and efficient for components detection, which use direct least square algorithm. However this method often perform poorly when different spectra are similar in shape. The basic rationale behind our WALOF method is that features in peak regions can provide additional information that can be used to refine the classification results of LOF based method. Fig. 1 provides an example explaining why and how WALOF works. If the 4th component does exist in the observation, the fitted curve (black in Fig. 1(a)) is likely to miss some of the peaks compared to the observation. Therefore, some of the intensity peaks in the 4th component should correlate well with the peaks in the residual spectrum, particularly in the regions outside the peak windows of the observation spectrum, where influences from other neighbouring peaks are small. As seen from Fig. 1(b) and (c), since the 4th component is a true component (though its LOF=0.0656), the most left peak in the residual component and that in the 4th components correlate well.