Table 8.

Precision, recall, F1-scores of five F–CNN models, and two baseline methods for detecting crop field and boundary from four channels (blue-green-red-NIR2) high-resolution satellite images.

	Boundary			Cropland
	Precision	Recall	F1-Score	Precision		Recall		F1-Score
	Training
Random Forest	0.84	0.66	0.74	0.95	0.98		0.96
FCN-DKConv6	0.88	0.86	0.87	0.95	0.95		0.95
U-Net	0.83	0.83	0.83	0.97	0.97		0.97
SegNet	0.84	0.84	0.84	0.97	0.97		0.97
DenseNet56	0.92	0.91	0.92	0.96	0.96		0.96
DenseNet67	0.93	0.92	0.92	0.98	0.98		0.98
DenseNet103	0.94	0.94	0.94	0.99	0.99		0.99
Test
Random Forest	0.49	0.22	0.30	0.87	0.96			0.92
FCN-DKConv6	0.61	0.6	0.61	0.91	0.92			0.91
U-Net	0.65	0.65	0.65	0.92	0.92			0.92
SegNet	0.71	0.71	0.71	0.92	0.93			0.92
DenseNet56	0.75	0.74	0.74	0.94	0.94			0.94
DenseNet67	0.77	0.75	0.76	0.94	0.94			0.94
DenseNet103	0.78	0.78	0.78	0.95	0.95			0.95