Inception residual attention network for remote sensing image super-resolution

杨利霞

Inception residual attention network for remote sensingimage super-resolution

7 O0 H& X# {! W: Y$ H. C* d
' P: a$ N; h2 sHow to enhance the spatial resolution for a remote sensing image is
+ g8 g' j5 A/ v( M" d9 a3 [. `an important issue that we face. Many image super-resolution (SR)
techniques have been proposed for this purpose and deep con
( b# d1 j$ \  {. b5 S' @( ?volutional neural network (CNN) is the most effective approach in
% b8 |# F1 m; jrecent years. However, we observe that most CNN-based SR meth
! p1 F& \3 X! _; c. g4 \& H' Lods treat low-frequency areas and high-frequency areas equally,
' ]! k5 J" i. ^1 |" Rhence hindering the recovery of high-frequency information. In this
paper, we propose a network named inception residual attention
" k. @3 Q* N  A! a: gnetwork (IRAN) to address this problem. Specifically, we propose
2 V8 g2 ]9 c  x6 D$ _* qa spatial attention module to make the network adaptively learn
the importance of different spatial areas, so as to pay more atten
tion to the areas with high-frequency information. Furthermore, we
! n5 r% p! N7 X0 K  Ypresent an inception module to fuse local multilevel features, so as
! j& i" V7 V) W7 zto provide richer information for reconstructing detailed textures. In
0 i8 M1 z  q2 S; ^2 K/ o) ~: Morder to evaluate the effectiveness of the proposed method, a large
9 Q: Q7 X& Z2 F4 n2 e4 @' l, Cnumber of experiments are performed on UCMerced-LandUse data
' E! E/ a# ?+ L6 `$ v* Y4 I8 gset and the results show that the proposed method is superior to
the current state-of-the-art methods in both visual effects and
& \* g, V2 d3 ?& W5 Dobjective indicators.

& Z' t# t( K, t$ x
5 H6 b* o( h- k0 V# q