Inception residual attention network for remote sensing image super-resolution

杨利霞

Inception residual attention network for remote sensingimage super-resolution

- M% \- [3 |% f$ ^

How to enhance the spatial resolution for a remote sensing image is
7 T( O* E4 e4 R, ^2 ~an important issue that we face. Many image super-resolution (SR)
7 J7 e* B- N+ r+ Stechniques have been proposed for this purpose and deep con
" @/ b% h0 G% X2 ovolutional neural network (CNN) is the most effective approach in
! L% E* z! k6 ?1 Y4 b& T5 Drecent years. However, we observe that most CNN-based SR meth
& p; X; W; e" o* c0 rods treat low-frequency areas and high-frequency areas equally,
3 z. D# H, E: W: ]hence hindering the recovery of high-frequency information. In this
# {: B# \9 S2 G- w1 D  m; Fpaper, we propose a network named inception residual attention
network (IRAN) to address this problem. Specifically, we propose
) r% p/ H  j- t% @$ k5 F) oa spatial attention module to make the network adaptively learn
& ^! N& e1 g4 [, i+ c/ Rthe importance of different spatial areas, so as to pay more atten
tion to the areas with high-frequency information. Furthermore, we
' U: u1 F; r/ A; Y& J  npresent an inception module to fuse local multilevel features, so as
to provide richer information for reconstructing detailed textures. In
, Q& \% G7 H$ i, worder to evaluate the effectiveness of the proposed method, a large
5 e) J  {9 y5 nnumber of experiments are performed on UCMerced-LandUse data
set and the results show that the proposed method is superior to
the current state-of-the-art methods in both visual effects and
  `: Q. x( \. A, O2 b3 }- Jobjective indicators.
/ e! C2 _1 _! k+ E
4 O# b% E- g( l: F5 p