Figure 8

Flowchart of the data preparation pipeline. The solid arrows denote the data flow and the dashed arrows denote the influence by the parameters. The raw input data was preprocessed (P1) to remove background offsets and noise, to filter out overexposed measurements, to cut the data into the appropriate spectral range and to normalize it. The data was then split 25 times into 80–20% DRB and validation batches (P2). The median of each spectral bin was calculated across all DRB measurements and subtracted from both DRB and validation sets (P3a and P3b). The DR (SDCM, PCA) was applied to the DRB (P4) set. Passthrough denotes that no DR was applied for the no DR data set. The results were used to project DRB and validation into the dimensional reduced space (P5a and P5b). The final sets were used as input for the classification pipelines. Generated with pgf v3.1.9a.