Semantic Segmentation of Remote Sensing Image Based on Convolutional Neural Network and Mask Generation

Table 4

Evaluation indicators of each method on the Potsdam dataset (%).

Category	Evaluation	Low vegetation	Buildings	Impervious surfaces	Cars	Trees
U-Net	Precision	75.3	52.3	73.8	69.4	79.4
	Recall	63.2	78.4	82.3	75.1	80.1
	IoU	52.5	45.7	63.6	56.2	66.2
CASIA3	Precision	83.2	72.5	85.5	75.6	82.5
	Recall	81.7	83.3	85.4	87.2	87.1
	IoU	70.2	63.3	74.6	68.9	78.8
HUSTW5	Precision	85.5	75.0	85.3	78.5	68.4
	Recall	82.3	82.2	86.5	87.6	87.3
	IoU	72.1	66.8	77.6	70.1	80.0
Proposed	Precision	85.7	88.6	87.9	86.4	89.3
	Recall	85.5	85.9	89.1	89.4	90.2
	IoU	74.3	68.6	79.4	72.1	82.2

The bold values represent the maximum values of evaluation index in the same column, such as precision, recall, and IoU and the index values of this paper are greater than those of other algorithms.