SCU-Net: Semantic Segmentation Network for Learning Channel Information on Remote Sensing Images

Table 3

Labeling results of SCU-net-102-A and DFCN121C for each class on the GF-2 dataset.

Model		Cultivated land	Forest land	Others	Water area	Building	Road
Predict pixels		22288550	92223235	199063	11547582	2982585	1830985
Predict pixels (TP)	SCU-net-102-A	15161346	89547666	113169	11183917	886270	1517840
	DFCN121C	14883805	89361908	117344	11255276	1999480	289870
Predict pixels (FP)	SCU-net-102-A	7127204	2675569	85894	363665	2096315	313145
	DFCN121C	7404745	2861327	81719	292306	983105	1541115
Precision	SCU-net-102-A	68.02%	97.10%	43.15%	96.85%	70.29%	82.90%
	DFCN121C	66.78%	96.90%	58.95%	97.47%	67.04%	84.17%
Recall	SCU-net-102-A	82.88%	93.41%	50.98%	92.15%	86.90%	72.87%
	DFCN121C	83.20%	93.16%	50.11%	90.71%	86.89%	70.87%
F1-score	SCU-net-102-A	0.747	0.952	0.467	0.944	0.777	0.776
	DFCN121C	0.741	0.949	0.541	0.939	0.757	0.769