Edge Prior Multilayer Segmentation Network Based on Bayesian Framework

Table 1

Confusion matrix of data set 1.

Method	Category	Building	Forest	Farmland	Road	Others	Average acc.
Meanshift+CRF	Buildings	75.20	8.54	0.43	8.34	7.49	70.33
	Forest	10.20	72.97	2.11	6.48	8.24
	Farmland	0.92	24.02	26.36	0.32	48.38
	Road	6.97	4.69	1.10	58.46	28.78
	Others	3.18	2.35	5.44	11.62	77.42
Meanshift+SVM	Buildings	74.34	8.15	1.14	5.89	10.47	66.32
	Forest	10.79	64.69	13.53	2.73	8.26
	Farmland	1.15	27.74	26.04	2.47	42.60
	Road	8.59	5.04	4.74	57.12	24.51
	Others	5.22	3.12	8.16	10.80	72.69
FCN8	Buildings	94.06	1.24	0.2	1.55	2.96	79.30
	Forest	18.12	71.47	1.26	2.77	6.37
	Farmland	3.50	12.39	51.34	7.16	25.61
	Road	10.26	2.75	0.6	61.11	25.29
	Others	5.18	1.51	0.68	6.17	86.45
DeepLab	Buildings	95.38	0.67	0.21	0.77	2.97	80.85
	Forest	18.33	74.57	0.25	1.23	5.61
	Farmland	2.29	16.42	48.44	3.02	29.83
	Road	11.57	3.43	0.44	53.26	31.30
	Others	5.20	1.17	0.16	2.68	90.79
FCN8+DT+the fully connected CRF	Buildings	95.32	0.70	0.28	0.84	2.86	80.90
	Forest	18.25	74.85	0.20	1.31	5.40
	Farmland	2.24	17.33	47.92	3.12	29.39
	Road	11.57	3.50	0.44	55.70	29.42
	Others	5.28	1.23	0.16	2.91	90.42
FCN8+HED+DT+the fully connected CRF	Buildings	95.41	0.70	0.19	0.84	2.86	80.94
	Forest	18.25	74.85	0.20	1.31	5.40
	Farmland	2.24	17.33	47.92	3.12	29.39
	Road	11.57	3.50	0.44	55.07	29.42
	Others	5.28	1.23	0.16	2.91	90.42
Pro. approach	Buildings	95.94	0.77	0.18	0.90	2.21	81.39
	Forest	18.03	75.46	0.23	1.20	5.09
	Farmland	2.21	16.84	49.33	3.53	28.08
	Road	11.23	3.45	0.51	56.78	28.02
	Others	4.98	1.14	0.16	3.46	90.27