Optimization Strategy of a Stacked Autoencoder and Deep Belief Network in a Hyperspectral Remote-Sensing Image Classification Model

Table 2

Confusion matrix for the SAE-SR classification accuracy of ROSIS-3 data.

	Asphalt	Bare ground	Gravel	Grassland	Metal sheet	Brick	Shadows	Trees	Total	Accuracy (%)
Asphalt	422	10	49	0	0	21	0	0	502	84.06
Bare ground	0	295	40	0	0	0	0	0	335	88.06
Gravel	19	30	401	0	0	8	0	0	458	87.55
Grassland	0	0	0	371	0	0	12	2	385	96.36
Metal sheet	0	0	4	0	741	17	0	0	762	97.24
Brick	16	14	1	0	0	358	0	0	389	92.03
Shadows	13	0	0	0	0	0	701	45	759	92.36
Trees	0	0	0	0	0	0	22	488	510	95.69
Total	470	349	495	371	741	404	735	535
Accuracy (%)	89.79	84.52	81.01	100	100	88.61	95.37	91.21

Overall accuracy = 3777/4100 = 92.12%; kappa coefficient = (0.9212 − 0.1353)/(1 − 0.1353) = 0.91.