Fig 1.
The proposed LSANet architecture for LULC in HSI.
Fig 2.
Illustration of standard GSA (2(a)), axial self-attention (2(b)), cross shaped attention (2(c)) and the proposed LS-attention (2(d)).
In 2(b), 2(c) and 2(d), the shadow area represents the input features split into different groups on which SA is conducted, and the yellow dot can directly interact with the token covered in the shadow region. Here, represent height, width and channel of the image, respectively.
Fig 3.
Illustration of (a) standard embedding in ViT, (b) proposed CPE, (c) PEG block.
Fig 4.
Details of the Pavia University (PU) dataset with samples and color coding.
Fig 5.
Details of the Indiana Pines (IP) dataset with samples and color coding.
Fig 6.
Details of the Salinas Valley (SV) dataset with samples samples and color coding.
Fig 7.
Details of the Botswana dataset with samples samples and color coding.
Table 1.
Performance comparison on the PU dataset.
Table 2.
Quantitative results comparison on the IP dataset.
Table 3.
Performance comparison on the Botswana dataset.
Table 4.
Performance comparison on the SV dataset.
Fig 8.
(a)GT (b)2D-CNN (c)3D-CNN (d)HybridSN (e)CSIL (f)SF (g) DSGSF (h)MF (i) SS1DSwin and (j) LSANet.
Fig 9.
Illustration of the visual map on IP dataset.
(a)GT (b)2D-CNN (c)3D-CNN (d)HybridSN (e)CSIL (f)SF (g) DSGSF (h)MF (i) SS1DSwin and (j) LSANet.
Fig 10.
(a)GT (b)2D-CNN (c)3D-CNN (d)HybridSN (e)CSIL (f)SF (g) DSGSF (h)MF (i) SS1DSwin and (j) LSANet.
Fig 11.
The visual map on Botswana dataset.
(a)GT (b)2D-CNN (c)3D-CNN (d)HybridSN (e)CSIL (f)SF (g) DSGSF (h)MF (i) SS1DSwin and (j) LSANet.
Fig 12.
Illustration of limpid size on IP, PU, SV and Botswana datasets.
Table 5.
Performance evaluation using different components.
Fig 13.
Illustration of the (a) learning rate and (b) batch size on the IP, PU, SV and Botswana datasets.
Table 6.
LSANet performance on IP dataset with CPE and zero padding.
Fig 14.
Training time comparison on PU, IP, SV, and Botswana datasets.
Table 7.
Performance comparison using different attention mechanisms.
Table 8.
Performance comparison using different encoding techniques.
Table 9.
Parameters and flops comparison.