U-Net: A Smart Application with Multidimensional Attention Network for Remote Sensing Images

Table 6

Quantitative evaluation results of building detection for different methods.

Data sets	Test area	IoU			Recall			Precision
		U-Net	CAR-UNet	MDAU-Net	U-Net	CAR-UNet	MDAU-Net	U-Net	CAR-UNet	MDAU-Net
Massachusetts building data set	1	69.74	68.96	76.95	83.46	80.23	86.46	88.48	91.34	96.32
	2	66.98	69.73	77.82	80.27	81.32	85.24	94.24	93.21	95.11
	3	73.26	74.22	80.93	81.64	78.43	88.72	93.31	90.23	96.68
	Mean	70.96	72.31	78.35	82.36	82.42	87.68	92.34	94.82	97.04
WHU data set	1	69.42	69.83	76.42	83.24	81.47	82.03	90.18	90.21	93.02
	2	68.33	65.24	73.92	80.66	82.56	87.49	89.32	91.37	92.83
	3	74.26	73.82	78.46	81.24	80.33	89.32	86.87	93.06	94.37
	Mean	71.53	72.41	77.97	82.21	83.26	90.63	88.94	93.12	95.68

Bold shows the values of IOU, recall, and precision on the whole dataset.