SAR Image Target Recognition Based on Improved Residual Attention Network

Table 1. Data category and number of samples
View table
Table 1. Data category and number of samples
Category Number of samples in training set Number of samples in test set
2S1 299 274
BMP2 232 196
BRDM-2 298 274
BTR-60 256 195
BTR-70 233 196
D7 299 274
T62 299 273
T72 232 196
ZIL131 299 274
ZSU-23_4 299 274

Table 2. Recognition rates of different algorithms
View table
Table 2. Recognition rates of different algorithms
Algorithm Number of parameters /10⁷ Recognition rate in training set Recognition rate in test set Time /s
ResNet50 2.5 99.4 97.9 1760
ResNet101 4.5 90.7 90.9 2426
Inception ResNetV2 5.5 88.1 94.5 7967

Table 3. Recognition rate of original residual attention network
View table
Table 3. Recognition rate of original residual attention network
Number of parameters /10⁷ Recognition rate in training set /% Recognition rate in test set /% Training time /s
3.2 99.5 98.9 4023

Table 4. Experimental results of different improvement stages
View table
Table 4. Experimental results of different improvement stages
Stage Number of parameters /10⁷ Recognition rate in training set /% Recognition rate in test set /% Training time /s
3 1.1 99.5 98.9
2 1.2 99.5 99.4
1 1.5 99.7 99.6 2850
0 3.2 99.5 98.9 4023

Table 5. Recognition results of different models
View table
Table 5. Recognition results of different models
Model Average recognition rate /%
Model S 99.6
CMNet model^[25] 99.3
Faster R-CNN model^[26] 99.1
A-ConvNets model 98.1
SVM model^[27] 90.0

Table 6. Occlusion recognition results
View table
Table 6. Occlusion recognition results
Noise ratio /% 5 15 20 30 35
Recognition rate /% 99 99 99 99 99

Table 7. Recognition results of model under salt and pepper noise
View table
Table 7. Recognition results of model under salt and pepper noise
Noise ratio /% 5 10 15
Recognition rate /% 96 88 82

Tools

Get Citation

Copy Citation Text

Baodai Shi, Qin Zhang, Yao Li, Yuhuan Li. SAR Image Target Recognition Based on Improved Residual Attention Network[J]. Laser & Optoelectronics Progress, 2021, 58(8): 0810008

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites