DSNet-Based Remote Sensing Image Semantic Segmentation Method

Layer	Input size	Operation	Output size
$X^{0,0}$	3×512×512	ReLU 32，Conv（3×3） BatchNorm	32×512×512
$X^{1,0}$	32×512×512	ReLU 64，Conv（3×3） BatchNorm	64×256×256
$X^{2,0}$	64×256×256	ReLU 128 Conv（3×3） BatchNorm	128×128×128
$X^{3,0}$	128×128×128	ReLU 256，Conv（3×3） BatchNorm	256×64×64
$X^{4,0}$	256×64×64	ReLU 512，Conv（3×3） BatchNorm	512×32×32
$X^{3,1}$	512×32×32	ReLU 256，Conv（3×3） BatchNorm	256×64×64
$X^{2,2}$	256×64×64	ReLU 128，Conv（3×3） BatchNorm	128×128×128
$X^{1,3}$	128×128×128	ReLU 64，Conv（3×3） BatchNorm	64×256×256
$X^{0,1}, X^{0,2} X^{0,3} X^{0,4}$	32×512×512	ReLU 32，Conv（3×3） BatchNorm	32×512×512

Table 2. Ablation experiment
View table
Table 2. Ablation experiment
Model PA MPA mIoU
DSNet+BCE+Adam 0.901 0.954 0.831
DSNet+FL+Dics+Adam 0.939 0.962 0.839
DSNet+FL+Dics+AdamW 0.941 0.977 0.849

Table 3. Comparison of model prediction and evaluation indexes
View table
Table 3. Comparison of model prediction and evaluation indexes
Model PA MPA mIoU
PSPNet 0.898 0.913 0.797
UNet 0.913 0.956 0.812
UNet++ 0.928 0.961 0.838
Proposed model 0.941 0.977 0.849

Table 4. Segmentation effect comparison of DSNet model
View table
Table 4. Segmentation effect comparison of DSNet model
Image Label PSPNet UNet UNet++ Proposed model
1
2
3
4

Tools

Get Citation

Copy Citation Text

Fangxing Shi, line Zhou, Daming Zhu, Zhitao Fu. DSNet-Based Remote Sensing Image Semantic Segmentation Method[J]. Laser & Optoelectronics Progress, 2023, 60(6): 0628002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites