Deep Belief Network for Feature Extraction of Urban Artificial Targets

Table 4

Confusion matrix of Hyspex data DBN-SR classification accuracy.

Class no.	1	2	3	4	5	6	Total	Accuracy (%)
1	155	11	0	0	13	0	179	86.59
2	0	164	0	0	0	8	172	95.35
3	0	0	238	0	0	0	238	1
4	0	0	0	131	0	14	145	90.34
5	0	0	3	6	136	21	166	81.93
6	0	0	0	0	14	131	145	90.34
Total	155	175	241	137	163	174	1045
Accuracy (%)	1	93.71	98.76	95.62	83.44	75.29
Overall accuracy = 955/1045 = 91.39%
Kappa = (997975 − 173163)/(1092025 − 173163) = 0.8976

(Class no.1: water; 2: vegetation; 3: cement road; 4: magmatic; 5: automobile; 6: curtain wall).