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

Table 4

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

	Asphalt	Bare ground	Gravel	Grassland	Metal sheet	Brick	Shadows	Trees	Total	Accuracy (%)
Asphalt	426	19	31	0	0	26	0	0	502	85.53
Bare ground	0	275	45	0	0	15	0	0	335	81.82
Gravel	45	31	375	0	0	35	0	0	458	84.05
Grassland	0	0	0	321	0	0	42	22	385	94.01
Metal sheet	0	0	0	0	710	52	0	0	762	96.31
Brick	21	26	9	0	0	333	0	0	389	86.98
Shadows	11	0	0	0	0	0	698	50	759	93.95
Trees	0	0	0	0	0	0	31	479	510	95.33
Total	503	351	460	321	710	461	771	551
Accuracy (%)	84.69	78.35	81.52	100	100	72.23	90.53	86.90

Overall accuracy = 3617/4100 = 88.22%; kappa coefficient = (0.8822 − 0.1363)/(1 − 0.1363) = 0.87.