Table 1.

Number and distribution of image classes in training and validation phases for the 10-way classifier

Class	In training set	Samples per iteration (feature reduction)	Samples per iteration (final training)	In validation set
Clear	26,160	5,000	5,000	2,887
Precip	25,951	5,000	5,000	2,897
Crystal	6,415	5,000	5,000	748
Phase	5,332	5,000	5,000	564
Precip & crystal	5,843	5,000	5,000	619
Precip & skin	3,748	3,748	3,748	466
Phase & crystal	1,790	1,790	1,790	201
Phase & precip	384	384	384	45
Skin	613	613	613	62
Junk	430	430	430	39