Figure 8.

Effect of the source data used to train the peak recognition models on the average recognition accuracy (Eq. 21) evaluated on the large dataset D. The reference library ${\mathbb{T}}_1$ is made of randomly chosen ICP pulses. The second training set, weighted sampling${\mathbb{T}}_2$, is made of an equal number of randomly selected pulses from the large dataset D and challenging pulses randomly extracted from the challenging dataset D'. Both datasets contain 3000 pulses