The 3D FTIR maps were evaluated using hierarchical cluster analysis (HCA), a multivariate statistical approach for classifying spectroscopic data. HCA permits the identification of regions within a sample structure based on their spectral response. Regions where the points display similar spectral responses demonstrate minimal intra-cluster spectral differences, while those with different spectral responses show maximal inter-cluster differences [23 (link)]. As the major vibrations characterising the investigated materials are within the 1800–950 cm−1 region of the IR spectrum, second derivative and vector normalisation were applied to this region to process raw spectral data for HCA. The spectra were smoothed over 17 points.
The optimal clustering algorithm was experimentally determined, using Euclidean distance as the measure between clusters. Ward’s method was used for clustering and construction of heterogeneity dendrograms. This method generates multiple partitions of the original image and considers all cluster (similar spectra) combinations using analysis of variance to assess the distance between clusters [51 (link)]. The number of clusters was determined based on technical data and the heterogeneity dendrogram. HCA was performed using OPUS v.7.5 software (Bruker Optik GmbH, Ettlingen, Germany).
The optimal clustering algorithm was experimentally determined, using Euclidean distance as the measure between clusters. Ward’s method was used for clustering and construction of heterogeneity dendrograms. This method generates multiple partitions of the original image and considers all cluster (similar spectra) combinations using analysis of variance to assess the distance between clusters [51 (link)]. The number of clusters was determined based on technical data and the heterogeneity dendrogram. HCA was performed using OPUS v.7.5 software (Bruker Optik GmbH, Ettlingen, Germany).
Free full text: Click here
