DeepLabV3_DHC: Semantic Segmentation of Urban Unmanned Aerial Vehicle Remote Sensing Image

Guowen Sun; Xiaobo Luo; Kunqiang Zhang

doi:10.3788/LOP230886

Laser & Optoelectronics Progress, Volume. 61, Issue 4, 0428005(2024)

DeepLabV3_DHC: Semantic Segmentation of Urban Unmanned Aerial Vehicle Remote Sensing Image

Guowen Sun¹, Xiaobo Luo^1、*, and Kunqiang Zhang²

¹College of Computer Sciences and Technology, Chongqing Engineering Research Center for Spatial Big Data Intelligent Technology, Chongqing University of Posts and Telecommunications, Chongqing 400065, China

²School of Information Engineering and Automation, Kunming University of Technology, Kunming 650500, Yunnan, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. DeepLabV3_DHC network structure diagram

Download full size

Fig. 2. Three groups of 3×3 dilated convolved receptive fields. (a) The dilated ratio is 1; (b) the dilated ratio is 2; (c) the dilated ratio is 3

Download full size

Fig. 3. Ordinary convolution and depth separable convolution. (a) Ordinary convolution; (b) depth separable convolution

Download full size

Fig. 4. Depthwise separable hybrid dilated convolution

Download full size

Fig. 5. Change in the number of dense upsampling channels

Download full size

Fig. 6. Mechanism of attention

Download full size

Fig. 7. Original image and corresponding label diagram

Download full size

Fig. 8. Original image and local histogram equalization

Download full size

Fig. 9. Visualization result chart. (a) Original image; (b) Label image; (c) DeepLabV3_ plus image segmentation; (d) DeepLabV3_ DHC image segmentation

Download full size

Table 1. Test set numerical evaluation

View table

Table 1. Test set numerical evaluation

Model

Backbone

MIOU /%

MPA /%

DeepLabV3_plus

DeepLabV3_DHC

MobilenetV2

MobilenetV3

Resnet101

Resnet152

Resnext101

Xception

MobilenetV2

MobilenetV3

Resnet101

Resnet152

Resnext101

Xception

66.58

63.51

73.67

75.58

75.68

78.22

71.73

70.68

78.10

76.19

78.27

80.99

76.63

74.13

84.97

86.37

85.99

86.48

80.65

79.49

85.81

86.64

86.79

90.09

Table 2. Comparison of parameter number and calculation amount

View table

Table 2. Comparison of parameter number and calculation amount

Model	Backbone	Parameters /10⁶	FLOPs /10⁹
DeepLabV3_plus	MobilenetV2	5.831	26.433
DeepLabV3_DHC	MobilenetV2	4.294	9.215
DeepLabV3_plus	Xception	54.709	83.420
DeepLabV3_DHC	Xception	51.513	65.195
DeepLabV3_plus	MobilenetV3	11.725	30.535
DeepLabV3_DHC	MobilenetV3	9.572	12.835
DeepLabV3_plus	Resnet101	59.354	199.408
DeepLabV3_DHC	Resnet101	56.181	181.863

Table 3. Different convolution values in the test set

View table

Table 3. Different convolution values in the test set

Convolution size

Dilated rate

MIOU /%

MPA /%

3×3

5×5

7×7

9×9

11×11

13×13

15×15

1，3，8

1，3，5

1，2，3

80.99

81.21

80.58

80.43

80.63

81.03

80.73

90.09

89.41

89.07

89.68

88.15

88.97

88.92

Table 4. Comparison of ablation experiments
View table
Table 4. Comparison of ablation experiments
Model MIOU /% MPA /%
DeepLabV3_DHC
（no DHC）
71.78 78.30
DeepLabV3_DHC
（no DUC）
70.83 82.83
DeepLabV3_DHC（no attention） 77.11 83.78
DeepLabV3_DHC 80.99 90.09

Tools

Get Citation

Copy Citation Text

Guowen Sun, Xiaobo Luo, Kunqiang Zhang. DeepLabV3_DHC: Semantic Segmentation of Urban Unmanned Aerial Vehicle Remote Sensing Image[J]. Laser & Optoelectronics Progress, 2024, 61(4): 0428005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites