A Novel THz Differential Spectral Clustering Recognition Method Based on t-SNE

Table 2

Model parameters for four samples.

Sample name	Spectral dimension	PCA dimension reduction	t-SNE dimension	Maximum number of iterations	Perplexity	Number of spectra in data set	Number of sample points
BeO sheet with defects	800	50	2	1000, 5000	30–100	45240	2000–10000
BeO sheet with zero defects	800	50	2	1000, 5000	100	45240	3000–10000
Al2O3 sheet with defects	800	50	2	1000, 5000	100	8400	8400
Monocrystalline silicon	800	50	2	1000, 5000	100	108800	5000–12000